]> git.saurik.com Git - apple/ld64.git/blob - src/ld/parsers/macho_relocatable_file.cpp
projects / apple / ld64.git / blob
? search:


776e695a3846fa229a9efd02ea30efb2691e7ff6
[apple/ld64.git] / src / ld / parsers / macho_relocatable_file.cpp
1 /* -*- mode: C++; c-basic-offset: 4; tab-width: 4 -*-
2 *
3 * Copyright (c) 2009-2010 Apple Inc. All rights reserved.
4 *
5 * @APPLE_LICENSE_HEADER_START@
6 *
7 * This file contains Original Code and/or Modifications of Original Code
8 * as defined in and that are subject to the Apple Public Source License
9 * Version 2.0 (the 'License'). You may not use this file except in
10 * compliance with the License. Please obtain a copy of the License at
11 * http://www.opensource.apple.com/apsl/ and read it before using this
12 * file.
13 *
14 * The Original Code and all software distributed under the License are
15 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
16 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
17 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
19 * Please see the License for the specific language governing rights and
20 * limitations under the License.
21 *
22 * @APPLE_LICENSE_HEADER_END@
23 */
24
25
26 #include <stdint.h>
27 #include <stdlib.h>
28 #include <math.h>
29 #include <unistd.h>
30 #include <fcntl.h>
31 #include <sys/param.h>
32 #include <sys/stat.h>
33 #include <sys/mman.h>
34
35 #include "MachOFileAbstraction.hpp"
36
37 #include "libunwind/DwarfInstructions.hpp"
38 #include "libunwind/AddressSpace.hpp"
39 #include "libunwind/Registers.hpp"
40
41 #include <vector>
42 #include <set>
43 #include <map>
44 #include <algorithm>
45 #include <type_traits>
46
47 #include "dwarf2.h"
48 #include "debugline.h"
49
50 #include "Architectures.hpp"
51 #include "Bitcode.hpp"
52 #include "ld.hpp"
53 #include "macho_relocatable_file.h"
54
55
56
57 extern void throwf(const char* format, ...) __attribute__ ((noreturn,format(printf, 1, 2)));
58 extern void warning(const char* format, ...) __attribute__((format(printf, 1, 2)));
59
60 namespace mach_o {
61 namespace relocatable {
62
63
64 // forward reference
65 template <typename A> class Parser;
66 template <typename A> class Atom;
67 template <typename A> class Section;
68 template <typename A> class CFISection;
69 template <typename A> class CUSection;
70
71 template <typename A>
72 class File : public ld::relocatable::File
73 {
74 public:
75 File(const char* p, time_t mTime, const uint8_t* content, ld::File::Ordinal ord) :
76 ld::relocatable::File(p,mTime,ord), _fileContent(content),
77 _sectionsArray(NULL), _atomsArray(NULL),
78 _sectionsArrayCount(0), _atomsArrayCount(0), _aliasAtomsArrayCount(0),
79 _debugInfoKind(ld::relocatable::File::kDebugInfoNone),
80 _dwarfTranslationUnitPath(NULL),
81 _dwarfDebugInfoSect(NULL), _dwarfDebugAbbrevSect(NULL),
82 _dwarfDebugLineSect(NULL), _dwarfDebugStringSect(NULL),
83 _objConstraint(ld::File::objcConstraintNone),
84 _swiftVersion(0),
85 _cpuSubType(0),
86 _minOSVersion(0),
87 _platform(0),
88 _canScatterAtoms(false),
89 _objcHasCategoryClassPropertiesField(false),
90 _srcKind(kSourceUnknown) { }
91 virtual ~File();
92
93 // overrides of ld::File
94 virtual bool forEachAtom(ld::File::AtomHandler&) const;
95 virtual bool justInTimeforEachAtom(const char* name, ld::File::AtomHandler&) const
96 { return false; }
97 virtual uint32_t minOSVersion() const { return _minOSVersion; }
98 virtual uint32_t platformLoadCommand() const { return _platform; }
99
100 // overrides of ld::relocatable::File
101 virtual ObjcConstraint objCConstraint() const { return _objConstraint; }
102 virtual bool objcHasCategoryClassPropertiesField() const
103 { return _objcHasCategoryClassPropertiesField; }
104 virtual uint32_t cpuSubType() const { return _cpuSubType; }
105 virtual DebugInfoKind debugInfo() const { return _debugInfoKind; }
106 virtual const std::vector<ld::relocatable::File::Stab>* stabs() const { return &_stabs; }
107 virtual bool canScatterAtoms() const { return _canScatterAtoms; }
108 virtual const char* translationUnitSource() const;
109 virtual LinkerOptionsList* linkerOptions() const { return &_linkerOptions; }
110 virtual uint8_t swiftVersion() const { return _swiftVersion; }
111 virtual ld::Bitcode* getBitcode() const { return _bitcode.get(); }
112 virtual SourceKind sourceKind() const { return _srcKind; }
113
114 const uint8_t* fileContent() { return _fileContent; }
115 private:
116 friend class Atom<A>;
117 friend class Section<A>;
118 friend class Parser<A>;
119 friend class CFISection<A>::OAS;
120
121 typedef typename A::P P;
122
123 const uint8_t* _fileContent;
124 Section<A>** _sectionsArray;
125 uint8_t* _atomsArray;
126 uint8_t* _aliasAtomsArray;
127 uint32_t _sectionsArrayCount;
128 uint32_t _atomsArrayCount;
129 uint32_t _aliasAtomsArrayCount;
130 std::vector<ld::Fixup> _fixups;
131 std::vector<ld::Atom::UnwindInfo> _unwindInfos;
132 std::vector<ld::Atom::LineInfo> _lineInfos;
133 std::vector<ld::relocatable::File::Stab>_stabs;
134 ld::relocatable::File::DebugInfoKind _debugInfoKind;
135 const char* _dwarfTranslationUnitPath;
136 const macho_section<P>* _dwarfDebugInfoSect;
137 const macho_section<P>* _dwarfDebugAbbrevSect;
138 const macho_section<P>* _dwarfDebugLineSect;
139 const macho_section<P>* _dwarfDebugStringSect;
140 ld::File::ObjcConstraint _objConstraint;
141 uint8_t _swiftVersion;
142 uint32_t _cpuSubType;
143 uint32_t _minOSVersion;
144 uint32_t _platform;
145 bool _canScatterAtoms;
146 bool _objcHasCategoryClassPropertiesField;
147 std::vector<std::vector<const char*> > _linkerOptions;
148 std::unique_ptr<ld::Bitcode> _bitcode;
149 SourceKind _srcKind;
150 };
151
152
153 template <typename A>
154 class Section : public ld::Section
155 {
156 public:
157 typedef typename A::P::uint_t pint_t;
158 typedef typename A::P P;
159 typedef typename A::P::E E;
160
161 virtual ~Section() { }
162 class File<A>& file() const { return _file; }
163 const macho_section<P>* machoSection() const { return _machOSection; }
164 uint32_t sectionNum(class Parser<A>&) const;
165 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr);
166 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeUnclassified; }
167 virtual bool dontDeadStrip() { return (this->_machOSection->flags() & S_ATTR_NO_DEAD_STRIP); }
168 virtual bool dontDeadStripIfReferencesLive() { return ( (this->_machOSection != NULL) && (this->_machOSection->flags() & S_ATTR_LIVE_SUPPORT) ); }
169 virtual Atom<A>* findAtomByAddress(pint_t addr) { return this->findContentAtomByAddress(addr, this->_beginAtoms, this->_endAtoms); }
170 virtual bool addFollowOnFixups() const { return ! _file.canScatterAtoms(); }
171 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer,
172 struct Parser<A>::LabelAndCFIBreakIterator& it,
173 const struct Parser<A>::CFI_CU_InfoArrays&) = 0;
174 virtual uint32_t computeAtomCount(class Parser<A>& parser,
175 struct Parser<A>::LabelAndCFIBreakIterator& it,
176 const struct Parser<A>::CFI_CU_InfoArrays&) = 0;
177 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&);
178 virtual bool addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>*);
179 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const { return 0; }
180 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
181 const ld::IndirectBindingTable& ind) const { return false; }
182 virtual bool ignoreLabel(const char* label) const { return false; }
183 static const char* makeSectionName(const macho_section<typename A::P>* s);
184
185 protected:
186 Section(File<A>& f, const macho_section<typename A::P>* s)
187 : ld::Section(makeSegmentName(s), makeSectionName(s), sectionType(s)),
188 _file(f), _machOSection(s), _beginAtoms(NULL), _endAtoms(NULL), _hasAliases(false) { }
189 Section(File<A>& f, const char* segName, const char* sectName, ld::Section::Type t, bool hidden=false)
190 : ld::Section(segName, sectName, t, hidden), _file(f), _machOSection(NULL),
191 _beginAtoms(NULL), _endAtoms(NULL), _hasAliases(false) { }
192
193
194 Atom<A>* findContentAtomByAddress(pint_t addr, class Atom<A>* start, class Atom<A>* end);
195 uint32_t x86_64PcRelOffset(uint8_t r_type);
196 void addLOH(class Parser<A>& parser, int kind, int count, const uint64_t addrs[]);
197 static const char* makeSegmentName(const macho_section<typename A::P>* s);
198 static bool readable(const macho_section<typename A::P>* s);
199 static bool writable(const macho_section<typename A::P>* s);
200 static bool exectuable(const macho_section<typename A::P>* s);
201 static ld::Section::Type sectionType(const macho_section<typename A::P>* s);
202
203 File<A>& _file;
204 const macho_section<P>* _machOSection;
205 class Atom<A>* _beginAtoms;
206 class Atom<A>* _endAtoms;
207 bool _hasAliases;
208 std::set<const class Atom<A>*> _altEntries;
209 };
210
211
212 template <typename A>
213 class CFISection : public Section<A>
214 {
215 public:
216 CFISection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
217 : Section<A>(f, s) { }
218 uint32_t cfiCount(Parser<A>& parser);
219
220 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeCFI; }
221 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&);
222 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&);
223 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&);
224 virtual bool addFollowOnFixups() const { return false; }
225
226
227 ///
228 /// ObjectFileAddressSpace is used as a template parameter to UnwindCursor for parsing
229 /// dwarf CFI information in an object file.
230 ///
231 class OAS
232 {
233 public:
234 typedef typename A::P::uint_t pint_t;
235 typedef typename A::P P;
236 typedef typename A::P::E E;
237 typedef typename A::P::uint_t sint_t;
238
239 OAS(CFISection<A>& ehFrameSection, const uint8_t* ehFrameBuffer) :
240 _ehFrameSection(ehFrameSection),
241 _ehFrameContent(ehFrameBuffer),
242 _ehFrameStartAddr(ehFrameSection.machoSection()->addr()),
243 _ehFrameEndAddr(ehFrameSection.machoSection()->addr()+ehFrameSection.machoSection()->size()) {}
244
245 uint8_t get8(pint_t addr) { return *((uint8_t*)mappedAddress(addr)); }
246 uint16_t get16(pint_t addr) { return E::get16(*((uint16_t*)mappedAddress(addr))); }
247 uint32_t get32(pint_t addr) { return E::get32(*((uint32_t*)mappedAddress(addr))); }
248 uint64_t get64(pint_t addr) { return E::get64(*((uint64_t*)mappedAddress(addr))); }
249 pint_t getP(pint_t addr) { return P::getP(*((pint_t*)mappedAddress(addr))); }
250 uint64_t getULEB128(pint_t& addr, pint_t end);
251 int64_t getSLEB128(pint_t& addr, pint_t end);
252 pint_t getEncodedP(pint_t& addr, pint_t end, uint8_t encoding);
253 private:
254 const void* mappedAddress(pint_t addr);
255
256 CFISection<A>& _ehFrameSection;
257 const uint8_t* _ehFrameContent;
258 pint_t _ehFrameStartAddr;
259 pint_t _ehFrameEndAddr;
260 };
261
262
263 typedef typename A::P::uint_t pint_t;
264 typedef libunwind::CFI_Atom_Info<OAS> CFI_Atom_Info;
265
266 void cfiParse(class Parser<A>& parser, uint8_t* buffer, CFI_Atom_Info cfiArray[], uint32_t& cfiCount, const pint_t cuStarts[], uint32_t cuCount);
267 bool needsRelocating();
268
269 static bool bigEndian();
270 private:
271 void addCiePersonalityFixups(class Parser<A>& parser, const CFI_Atom_Info* cieInfo);
272 static void warnFunc(void* ref, uint64_t funcAddr, const char* msg);
273 };
274
275
276 template <typename A>
277 class CUSection : public Section<A>
278 {
279 public:
280 CUSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
281 : Section<A>(f, s) { }
282
283 typedef typename A::P::uint_t pint_t;
284 typedef typename A::P P;
285 typedef typename A::P::E E;
286
287 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&) { return 0; }
288 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&) { return 0; }
289 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&);
290 virtual bool addFollowOnFixups() const { return false; }
291
292 struct Info {
293 pint_t functionStartAddress;
294 uint32_t functionSymbolIndex;
295 uint32_t rangeLength;
296 uint32_t compactUnwindInfo;
297 const char* personality;
298 pint_t lsdaAddress;
299 Atom<A>* function;
300 Atom<A>* lsda;
301 };
302
303 uint32_t count();
304 void parse(class Parser<A>& parser, uint32_t cnt, Info array[]);
305 static bool encodingMeansUseDwarf(compact_unwind_encoding_t enc);
306
307
308 private:
309
310 const char* personalityName(class Parser<A>& parser, const macho_relocation_info<P>* reloc);
311
312 static int infoSorter(const void* l, const void* r);
313
314 };
315
316
317 template <typename A>
318 class TentativeDefinitionSection : public Section<A>
319 {
320 public:
321 TentativeDefinitionSection(Parser<A>& parser, File<A>& f)
322 : Section<A>(f, "__DATA", "__comm/tent", ld::Section::typeTentativeDefs) {}
323
324 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeZeroFill; }
325 virtual bool addFollowOnFixups() const { return false; }
326 virtual Atom<A>* findAtomByAddress(typename A::P::uint_t addr) { throw "TentativeDefinitionSection::findAtomByAddress() should never be called"; }
327 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it,
328 const struct Parser<A>::CFI_CU_InfoArrays&);
329 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer,
330 struct Parser<A>::LabelAndCFIBreakIterator& it,
331 const struct Parser<A>::CFI_CU_InfoArrays&);
332 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&) {}
333 private:
334 typedef typename A::P::uint_t pint_t;
335 typedef typename A::P P;
336 };
337
338
339 template <typename A>
340 class AbsoluteSymbolSection : public Section<A>
341 {
342 public:
343 AbsoluteSymbolSection(Parser<A>& parser, File<A>& f)
344 : Section<A>(f, "__DATA", "__abs", ld::Section::typeAbsoluteSymbols, true) {}
345
346 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeUnclassified; }
347 virtual bool dontDeadStrip() { return false; }
348 virtual ld::Atom::Alignment alignmentForAddress(typename A::P::uint_t addr) { return ld::Atom::Alignment(0); }
349 virtual bool addFollowOnFixups() const { return false; }
350 virtual Atom<A>* findAtomByAddress(typename A::P::uint_t addr) { throw "AbsoluteSymbolSection::findAtomByAddress() should never be called"; }
351 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it,
352 const struct Parser<A>::CFI_CU_InfoArrays&);
353 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer,
354 struct Parser<A>::LabelAndCFIBreakIterator& it,
355 const struct Parser<A>::CFI_CU_InfoArrays&);
356 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&) {}
357 virtual Atom<A>* findAbsAtomForValue(typename A::P::uint_t);
358
359 private:
360 typedef typename A::P::uint_t pint_t;
361 typedef typename A::P P;
362 };
363
364
365 template <typename A>
366 class SymboledSection : public Section<A>
367 {
368 public:
369 SymboledSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s);
370 virtual ld::Atom::ContentType contentType() { return _type; }
371 virtual bool dontDeadStrip();
372 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it,
373 const struct Parser<A>::CFI_CU_InfoArrays&);
374 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer,
375 struct Parser<A>::LabelAndCFIBreakIterator& it,
376 const struct Parser<A>::CFI_CU_InfoArrays&);
377 protected:
378 typedef typename A::P::uint_t pint_t;
379 typedef typename A::P P;
380
381 ld::Atom::ContentType _type;
382 };
383
384
385 template <typename A>
386 class TLVDefsSection : public SymboledSection<A>
387 {
388 public:
389 TLVDefsSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s) :
390 SymboledSection<A>(parser, f, s) { }
391
392 typedef typename A::P::uint_t pint_t;
393
394 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
395
396 private:
397
398 };
399
400
401 template <typename A>
402 class ImplicitSizeSection : public Section<A>
403 {
404 public:
405 ImplicitSizeSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
406 : Section<A>(f, s) { }
407 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&);
408 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&);
409 protected:
410 typedef typename A::P::uint_t pint_t;
411 typedef typename A::P P;
412
413 virtual bool addFollowOnFixups() const { return false; }
414 virtual const char* unlabeledAtomName(Parser<A>& parser, pint_t addr) = 0;
415 virtual ld::Atom::SymbolTableInclusion symbolTableInclusion();
416 virtual pint_t elementSizeAtAddress(pint_t addr) = 0;
417 virtual ld::Atom::Scope scopeAtAddress(Parser<A>& parser, pint_t addr) { return ld::Atom::scopeLinkageUnit; }
418 virtual bool useElementAt(Parser<A>& parser,
419 struct Parser<A>::LabelAndCFIBreakIterator& it, pint_t addr) = 0;
420 virtual ld::Atom::Definition definition() { return ld::Atom::definitionRegular; }
421 virtual ld::Atom::Combine combine(Parser<A>& parser, pint_t addr) = 0;
422 virtual bool ignoreLabel(const char* label) const { return (label[0] == 'L'); }
423 };
424
425
426 template <typename A>
427 class FixedSizeSection : public ImplicitSizeSection<A>
428 {
429 public:
430 FixedSizeSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
431 : ImplicitSizeSection<A>(parser, f, s) { }
432 protected:
433 typedef typename A::P::uint_t pint_t;
434 typedef typename A::P P;
435 typedef typename A::P::E E;
436
437 virtual bool useElementAt(Parser<A>& parser,
438 struct Parser<A>::LabelAndCFIBreakIterator& it, pint_t addr)
439 { return true; }
440 };
441
442
443 template <typename A>
444 class Literal4Section : public FixedSizeSection<A>
445 {
446 public:
447 Literal4Section(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
448 : FixedSizeSection<A>(parser, f, s) {}
449 protected:
450 typedef typename A::P::uint_t pint_t;
451 typedef typename A::P P;
452
453 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(2); }
454 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "4-byte-literal"; }
455 virtual pint_t elementSizeAtAddress(pint_t addr) { return 4; }
456 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
457 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
458 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
459 const ld::IndirectBindingTable& ind) const;
460 virtual bool ignoreLabel(const char* label) const;
461 };
462
463 template <typename A>
464 class Literal8Section : public FixedSizeSection<A>
465 {
466 public:
467 Literal8Section(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
468 : FixedSizeSection<A>(parser, f, s) {}
469 protected:
470 typedef typename A::P::uint_t pint_t;
471 typedef typename A::P P;
472
473 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(3); }
474 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "8-byte-literal"; }
475 virtual pint_t elementSizeAtAddress(pint_t addr) { return 8; }
476 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
477 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
478 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
479 const ld::IndirectBindingTable& ind) const;
480 virtual bool ignoreLabel(const char* label) const;
481 };
482
483 template <typename A>
484 class Literal16Section : public FixedSizeSection<A>
485 {
486 public:
487 Literal16Section(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
488 : FixedSizeSection<A>(parser, f, s) {}
489 protected:
490 typedef typename A::P::uint_t pint_t;
491 typedef typename A::P P;
492
493 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(4); }
494 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "16-byte-literal"; }
495 virtual pint_t elementSizeAtAddress(pint_t addr) { return 16; }
496 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
497 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
498 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
499 const ld::IndirectBindingTable& ind) const;
500 virtual bool ignoreLabel(const char* label) const;
501 };
502
503
504 template <typename A>
505 class NonLazyPointerSection : public FixedSizeSection<A>
506 {
507 public:
508 NonLazyPointerSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
509 : FixedSizeSection<A>(parser, f, s) {}
510 protected:
511 typedef typename A::P::uint_t pint_t;
512 typedef typename A::P P;
513
514 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&);
515 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeNonLazyPointer; }
516 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
517 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "non_lazy_ptr"; }
518 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
519 virtual ld::Atom::Scope scopeAtAddress(Parser<A>& parser, pint_t addr);
520 virtual ld::Atom::Combine combine(Parser<A>&, pint_t);
521 virtual bool ignoreLabel(const char* label) const { return true; }
522 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
523 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
524 const ld::IndirectBindingTable& ind) const;
525
526 private:
527 static const char* targetName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind);
528 static ld::Fixup::Kind fixupKind();
529 };
530
531 template <typename A>
532 class TLVPointerSection : public FixedSizeSection<A>
533 {
534 public:
535 TLVPointerSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
536 : FixedSizeSection<A>(parser, f, s) {}
537 protected:
538 typedef typename A::P::uint_t pint_t;
539 typedef typename A::P P;
540
541 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&);
542 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeTLVPointer; }
543 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
544 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "tlv_lazy_ptr"; }
545 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
546 virtual ld::Atom::Combine combine(Parser<A>&, pint_t);
547 virtual bool ignoreLabel(const char* label) const { return true; }
548 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
549 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
550 const ld::IndirectBindingTable& ind) const;
551
552 private:
553 static const char* targetName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind, bool* isStatic);
554 };
555
556
557 template <typename A>
558 class CFStringSection : public FixedSizeSection<A>
559 {
560 public:
561 CFStringSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
562 : FixedSizeSection<A>(parser, f, s) {}
563 protected:
564 typedef typename A::P::uint_t pint_t;
565
566 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
567 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "CFString"; }
568 virtual pint_t elementSizeAtAddress(pint_t addr) { return 4*sizeof(pint_t); }
569 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndReferences; }
570 virtual bool ignoreLabel(const char* label) const { return true; }
571 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
572 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
573 const ld::IndirectBindingTable& ind) const;
574 private:
575 enum ContentType { contentUTF8, contentUTF16, contentUnknown };
576 static const uint8_t* targetContent(const class Atom<A>* atom, const ld::IndirectBindingTable& ind,
577 ContentType* ct, unsigned int* count);
578 };
579
580
581 template <typename A>
582 class ObjC1ClassSection : public FixedSizeSection<A>
583 {
584 public:
585 ObjC1ClassSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
586 : FixedSizeSection<A>(parser, f, s) {}
587 protected:
588 typedef typename A::P::uint_t pint_t;
589 typedef typename A::P P;
590 typedef typename A::P::E E;
591
592 virtual ld::Atom::Scope scopeAtAddress(Parser<A>& , pint_t ) { return ld::Atom::scopeGlobal; }
593 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(2); }
594 virtual const char* unlabeledAtomName(Parser<A>&, pint_t);
595 virtual ld::Atom::SymbolTableInclusion symbolTableInclusion() { return ld::Atom::symbolTableIn; }
596 virtual pint_t elementSizeAtAddress(pint_t addr);
597 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineNever; }
598 virtual bool ignoreLabel(const char* label) const { return true; }
599 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
600 { return 0; }
601 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
602 const ld::IndirectBindingTable& ind) const { return false; }
603 virtual bool addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>*);
604 };
605
606
607 template <typename A>
608 class ObjC2ClassRefsSection : public FixedSizeSection<A>
609 {
610 public:
611 ObjC2ClassRefsSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
612 : FixedSizeSection<A>(parser, f, s) {}
613 protected:
614 typedef typename A::P::uint_t pint_t;
615
616 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
617 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "objc-class-ref"; }
618 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
619 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndReferences; }
620 virtual bool ignoreLabel(const char* label) const { return true; }
621 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
622 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
623 const ld::IndirectBindingTable& ind) const;
624 private:
625 const char* targetClassName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
626 };
627
628
629 template <typename A>
630 class ObjC2CategoryListSection : public FixedSizeSection<A>
631 {
632 public:
633 ObjC2CategoryListSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
634 : FixedSizeSection<A>(parser, f, s) {}
635 protected:
636 typedef typename A::P::uint_t pint_t;
637
638 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
639 virtual ld::Atom::Scope scopeAtAddress(Parser<A>& parser, pint_t addr) { return ld::Atom::scopeTranslationUnit; }
640 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "objc-cat-list"; }
641 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
642 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineNever; }
643 virtual bool ignoreLabel(const char* label) const { return true; }
644 private:
645 const char* targetClassName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
646 };
647
648
649 template <typename A>
650 class PointerToCStringSection : public FixedSizeSection<A>
651 {
652 public:
653 PointerToCStringSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
654 : FixedSizeSection<A>(parser, f, s) {}
655 protected:
656 typedef typename A::P::uint_t pint_t;
657
658 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
659 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "pointer-to-literal-cstring"; }
660 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
661 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndReferences; }
662 virtual bool ignoreLabel(const char* label) const { return true; }
663 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
664 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
665 const ld::IndirectBindingTable& ind) const;
666 virtual const char* targetCString(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
667 };
668
669
670 template <typename A>
671 class Objc1ClassReferences : public PointerToCStringSection<A>
672 {
673 public:
674 Objc1ClassReferences(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
675 : PointerToCStringSection<A>(parser, f, s) {}
676
677 typedef typename A::P::uint_t pint_t;
678 typedef typename A::P P;
679
680 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "pointer-to-literal-objc-class-name"; }
681 virtual bool addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>*);
682 virtual const char* targetCString(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
683 };
684
685
686 template <typename A>
687 class CStringSection : public ImplicitSizeSection<A>
688 {
689 public:
690 CStringSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
691 : ImplicitSizeSection<A>(parser, f, s) {}
692 protected:
693 typedef typename A::P::uint_t pint_t;
694 typedef typename A::P P;
695
696 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeCString; }
697 virtual Atom<A>* findAtomByAddress(pint_t addr);
698 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "cstring"; }
699 virtual pint_t elementSizeAtAddress(pint_t addr);
700 virtual bool ignoreLabel(const char* label) const;
701 virtual bool useElementAt(Parser<A>& parser,
702 struct Parser<A>::LabelAndCFIBreakIterator& it, pint_t addr);
703 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
704 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
705 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
706 const ld::IndirectBindingTable& ind) const;
707
708 };
709
710
711 template <typename A>
712 class UTF16StringSection : public SymboledSection<A>
713 {
714 public:
715 UTF16StringSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
716 : SymboledSection<A>(parser, f, s) {}
717 protected:
718 typedef typename A::P::uint_t pint_t;
719 typedef typename A::P P;
720
721 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
722 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
723 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
724 const ld::IndirectBindingTable& ind) const;
725 };
726
727
728 //
729 // Atoms in mach-o files
730 //
731 template <typename A>
732 class Atom : public ld::Atom
733 {
734 public:
735 // overrides of ld::Atom
736 virtual const ld::File* file() const;
737 virtual const char* translationUnitSource() const
738 { return sect().file().translationUnitSource(); }
739 virtual const char* name() const { return _name; }
740 virtual uint64_t size() const { return _size; }
741 virtual uint64_t objectAddress() const { return _objAddress; }
742 virtual void copyRawContent(uint8_t buffer[]) const;
743 virtual const uint8_t* rawContentPointer() const { return contentPointer(); }
744 virtual unsigned long contentHash(const ld::IndirectBindingTable& ind) const
745 { if ( _hash == 0 ) _hash = sect().contentHash(this, ind); return _hash; }
746 virtual bool canCoalesceWith(const ld::Atom& rhs, const ld::IndirectBindingTable& ind) const
747 { return sect().canCoalesceWith(this, rhs, ind); }
748 virtual ld::Fixup::iterator fixupsBegin() const { return &machofile()._fixups[_fixupsStartIndex]; }
749 virtual ld::Fixup::iterator fixupsEnd() const { return &machofile()._fixups[_fixupsStartIndex+_fixupsCount]; }
750 virtual ld::Atom::UnwindInfo::iterator beginUnwind() const { return &machofile()._unwindInfos[_unwindInfoStartIndex]; }
751 virtual ld::Atom::UnwindInfo::iterator endUnwind() const { return &machofile()._unwindInfos[_unwindInfoStartIndex+_unwindInfoCount]; }
752 virtual ld::Atom::LineInfo::iterator beginLineInfo() const{ return &machofile()._lineInfos[_lineInfoStartIndex]; }
753 virtual ld::Atom::LineInfo::iterator endLineInfo() const { return &machofile()._lineInfos[_lineInfoStartIndex+_lineInfoCount]; }
754 virtual void setFile(const ld::File* f);
755
756 private:
757
758 enum { kFixupStartIndexBits = 32,
759 kLineInfoStartIndexBits = 32,
760 kUnwindInfoStartIndexBits = 24,
761 kFixupCountBits = 24,
762 kLineInfoCountBits = 12,
763 kUnwindInfoCountBits = 4
764 }; // must sum to 128
765
766 public:
767 // methods for all atoms from mach-o object file
768 Section<A>& sect() const { return (Section<A>&)section(); }
769 File<A>& machofile() const { return ((Section<A>*)(this->_section))->file(); }
770 void setFixupsRange(uint32_t s, uint32_t c);
771 void setUnwindInfoRange(uint32_t s, uint32_t c);
772 void extendUnwindInfoRange();
773 void setLineInfoRange(uint32_t s, uint32_t c);
774 bool roomForMoreLineInfoCount() { return (_lineInfoCount < ((1<<kLineInfoCountBits)-1)); }
775 void incrementLineInfoCount() { assert(roomForMoreLineInfoCount()); ++_lineInfoCount; }
776 void incrementFixupCount() { if (_fixupsCount == ((1 << kFixupCountBits)-1))
777 throwf("too may fixups in %s", name()); ++_fixupsCount; }
778 const uint8_t* contentPointer() const;
779 uint32_t fixupCount() const { return _fixupsCount; }
780 void verifyAlignment(const macho_section<typename A::P>&) const;
781
782 typedef typename A::P P;
783 typedef typename A::P::E E;
784 typedef typename A::P::uint_t pint_t;
785 // constuct via all attributes
786 Atom(Section<A>& sct, const char* nm, pint_t addr, uint64_t sz,
787 ld::Atom::Definition d, ld::Atom::Combine c, ld::Atom::Scope s,
788 ld::Atom::ContentType ct, ld::Atom::SymbolTableInclusion i,
789 bool dds, bool thumb, bool al, ld::Atom::Alignment a)
790 : ld::Atom((ld::Section&)sct, d, c, s, ct, i, dds, thumb, al, a),
791 _size(sz), _objAddress(addr), _name(nm), _hash(0),
792 _fixupsStartIndex(0), _lineInfoStartIndex(0),
793 _unwindInfoStartIndex(0), _fixupsCount(0),
794 _lineInfoCount(0), _unwindInfoCount(0) { }
795 // construct via symbol table entry
796 Atom(Section<A>& sct, Parser<A>& parser, const macho_nlist<P>& sym,
797 uint64_t sz, bool alias=false)
798 : ld::Atom((ld::Section&)sct, parser.definitionFromSymbol(sym),
799 parser.combineFromSymbol(sym), parser.scopeFromSymbol(sym),
800 parser.resolverFromSymbol(sym) ? ld::Atom::typeResolver : sct.contentType(),
801 parser.inclusionFromSymbol(sym),
802 (parser.dontDeadStripFromSymbol(sym) && !sct.dontDeadStripIfReferencesLive()) || sct.dontDeadStrip(),
803 parser.isThumbFromSymbol(sym), alias,
804 sct.alignmentForAddress(sym.n_value())),
805 _size(sz), _objAddress(sym.n_value()),
806 _name(parser.nameFromSymbol(sym)), _hash(0),
807 _fixupsStartIndex(0), _lineInfoStartIndex(0),
808 _unwindInfoStartIndex(0), _fixupsCount(0),
809 _lineInfoCount(0), _unwindInfoCount(0) {
810 // <rdar://problem/6783167> support auto-hidden weak symbols
811 if ( _scope == ld::Atom::scopeGlobal &&
812 (sym.n_desc() & (N_WEAK_DEF|N_WEAK_REF)) == (N_WEAK_DEF|N_WEAK_REF) )
813 this->setAutoHide();
814 this->verifyAlignment(*sct.machoSection());
815 if ( sct.dontDeadStripIfReferencesLive() )
816 this->setDontDeadStripIfReferencesLive();
817 }
818
819 private:
820 friend class Parser<A>;
821 friend class Section<A>;
822 friend class CStringSection<A>;
823 friend class AbsoluteSymbolSection<A>;
824
825 pint_t _size;
826 pint_t _objAddress;
827 const char* _name;
828 mutable unsigned long _hash;
829
830 uint64_t _fixupsStartIndex : kFixupStartIndexBits,
831 _lineInfoStartIndex : kLineInfoStartIndexBits,
832 _unwindInfoStartIndex : kUnwindInfoStartIndexBits,
833 _fixupsCount : kFixupCountBits,
834 _lineInfoCount : kLineInfoCountBits,
835 _unwindInfoCount : kUnwindInfoCountBits;
836
837 static std::map<const ld::Atom*, const ld::File*> _s_fileOverride;
838 };
839
840 template <typename A>
841 std::map<const ld::Atom*, const ld::File*> Atom<A>::_s_fileOverride;
842
843 template <typename A>
844 void Atom<A>::setFile(const ld::File* f) {
845 _s_fileOverride[this] = f;
846 }
847
848 template <typename A>
849 const ld::File* Atom<A>::file() const
850 {
851 std::map<const ld::Atom*, const ld::File*>::iterator pos = _s_fileOverride.find(this);
852 if ( pos != _s_fileOverride.end() )
853 return pos->second;
854
855 return &sect().file();
856 }
857
858 template <typename A>
859 void Atom<A>::setFixupsRange(uint32_t startIndex, uint32_t count)
860 {
861 if ( count >= (1 << kFixupCountBits) )
862 throwf("too many fixups in function %s", this->name());
863 if ( startIndex >= (1 << kFixupStartIndexBits) )
864 throwf("too many fixups in file");
865 assert(((startIndex+count) <= sect().file()._fixups.size()) && "fixup index out of range");
866 _fixupsStartIndex = startIndex;
867 _fixupsCount = count;
868 }
869
870 template <typename A>
871 void Atom<A>::setUnwindInfoRange(uint32_t startIndex, uint32_t count)
872 {
873 if ( count >= (1 << kUnwindInfoCountBits) )
874 throwf("too many compact unwind infos in function %s", this->name());
875 if ( startIndex >= (1 << kUnwindInfoStartIndexBits) )
876 throwf("too many compact unwind infos (%d) in file", startIndex);
877 assert((startIndex+count) <= sect().file()._unwindInfos.size() && "unwindinfo index out of range");
878 _unwindInfoStartIndex = startIndex;
879 _unwindInfoCount = count;
880 }
881
882 template <typename A>
883 void Atom<A>::extendUnwindInfoRange()
884 {
885 if ( _unwindInfoCount+1 >= (1 << kUnwindInfoCountBits) )
886 throwf("too many compact unwind infos in function %s", this->name());
887 _unwindInfoCount += 1;
888 }
889
890 template <typename A>
891 void Atom<A>::setLineInfoRange(uint32_t startIndex, uint32_t count)
892 {
893 assert((count < (1 << kLineInfoCountBits)) && "too many line infos");
894 assert((startIndex+count) < sect().file()._lineInfos.size() && "line info index out of range");
895 _lineInfoStartIndex = startIndex;
896 _lineInfoCount = count;
897 }
898
899 template <typename A>
900 const uint8_t* Atom<A>::contentPointer() const
901 {
902 const macho_section<P>* sct = this->sect().machoSection();
903 if ( this->_objAddress > sct->addr() + sct->size() )
904 throwf("malformed .o file, symbol has address 0x%0llX which is outside range of its section", (uint64_t)this->_objAddress);
905 uint32_t fileOffset = sct->offset() - sct->addr() + this->_objAddress;
906 return this->sect().file().fileContent()+fileOffset;
907 }
908
909
910 template <typename A>
911 void Atom<A>::copyRawContent(uint8_t buffer[]) const
912 {
913 // copy base bytes
914 if ( this->contentType() == ld::Atom::typeZeroFill ) {
915 bzero(buffer, _size);
916 }
917 else if ( _size != 0 ) {
918 memcpy(buffer, this->contentPointer(), _size);
919 }
920 }
921
922 template <>
923 void Atom<arm>::verifyAlignment(const macho_section<P>&) const
924 {
925 if ( (this->section().type() == ld::Section::typeCode) && ! isThumb() ) {
926 if ( ((_objAddress % 4) != 0) || (this->alignment().powerOf2 < 2) )
927 warning("ARM function not 4-byte aligned: %s from %s", this->name(), this->file()->path());
928 }
929 }
930
931 #if SUPPORT_ARCH_arm64
932 template <>
933 void Atom<arm64>::verifyAlignment(const macho_section<P>& sect) const
934 {
935 if ( (this->section().type() == ld::Section::typeCode) && (sect.size() != 0) ) {
936 if ( ((_objAddress % 4) != 0) || (this->alignment().powerOf2 < 2) )
937 warning("arm64 function not 4-byte aligned: %s from %s", this->name(), this->file()->path());
938 }
939 }
940 #endif
941
942
943 template <typename A>
944 void Atom<A>::verifyAlignment(const macho_section<P>&) const
945 {
946 }
947
948
949 class AliasAtom : public ld::Atom
950 {
951 public:
952 AliasAtom(const char* name, bool hidden, const ld::File* file, const char* aliasOfName) :
953 ld::Atom(_s_section, ld::Atom::definitionRegular, ld::Atom::combineNever,
954 (hidden ? ld::Atom::scopeLinkageUnit : ld::Atom::scopeGlobal),
955 ld::Atom::typeUnclassified, ld::Atom::symbolTableIn,
956 false, false, true, 0),
957 _file(file),
958 _name(name),
959 _fixup(0, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, ld::Fixup::bindingByNameUnbound, aliasOfName) { }
960
961 virtual const ld::File* file() const { return _file; }
962 virtual const char* translationUnitSource() const
963 { return NULL; }
964 virtual const char* name() const { return _name; }
965 virtual uint64_t size() const { return 0; }
966 virtual uint64_t objectAddress() const { return 0; }
967 virtual void copyRawContent(uint8_t buffer[]) const { }
968 virtual ld::Fixup::iterator fixupsBegin() const { return &((ld::Fixup*)&_fixup)[0]; }
969 virtual ld::Fixup::iterator fixupsEnd() const { return &((ld::Fixup*)&_fixup)[1]; }
970
971 private:
972 static ld::Section _s_section;
973
974 const ld::File* _file;
975 const char* _name;
976 ld::Fixup _fixup;
977 };
978
979 ld::Section AliasAtom::_s_section("__LD", "__aliases", ld::Section::typeTempAlias, true);
980
981
982 template <typename A>
983 class Parser
984 {
985 public:
986 static bool validFile(const uint8_t* fileContent, bool subtypeMustMatch=false,
987 cpu_subtype_t subtype=0);
988 static const char* fileKind(const uint8_t* fileContent);
989 static Options::Platform findPlatform(const macho_header<typename A::P>* header);
990 static bool hasObjC2Categories(const uint8_t* fileContent);
991 static bool hasObjC1Categories(const uint8_t* fileContent);
992 static bool getNonLocalSymbols(const uint8_t* fileContnet, std::vector<const char*> &syms);
993 static ld::relocatable::File* parse(const uint8_t* fileContent, uint64_t fileLength,
994 const char* path, time_t modTime, ld::File::Ordinal ordinal,
995 const ParserOptions& opts) {
996 Parser p(fileContent, fileLength, path, modTime,
997 ordinal, opts.warnUnwindConversionProblems,
998 opts.keepDwarfUnwind, opts.forceDwarfConversion,
999 opts.neverConvertDwarf, opts.verboseOptimizationHints,
1000 opts.ignoreMismatchPlatform);
1001 return p.parse(opts);
1002 }
1003
1004 typedef typename A::P P;
1005 typedef typename A::P::E E;
1006 typedef typename A::P::uint_t pint_t;
1007
1008 struct SourceLocation {
1009 SourceLocation() {}
1010 SourceLocation(Atom<A>* a, uint32_t o) : atom(a), offsetInAtom(o) {}
1011 Atom<A>* atom;
1012 uint32_t offsetInAtom;
1013 };
1014
1015 struct TargetDesc {
1016 Atom<A>* atom;
1017 const char* name; // only used if targetAtom is NULL
1018 int64_t addend;
1019 bool weakImport; // only used if targetAtom is NULL
1020 };
1021
1022 struct FixupInAtom {
1023 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, Atom<A>* target) :
1024 fixup(src.offsetInAtom, c, k, target), atom(src.atom) { src.atom->incrementFixupCount(); }
1025
1026 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, ld::Fixup::TargetBinding b, Atom<A>* target) :
1027 fixup(src.offsetInAtom, c, k, b, target), atom(src.atom) { src.atom->incrementFixupCount(); }
1028
1029 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, bool wi, const char* name) :
1030 fixup(src.offsetInAtom, c, k, wi, name), atom(src.atom) { src.atom->incrementFixupCount(); }
1031
1032 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, ld::Fixup::TargetBinding b, const char* name) :
1033 fixup(src.offsetInAtom, c, k, b, name), atom(src.atom) { src.atom->incrementFixupCount(); }
1034
1035 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, uint64_t addend) :
1036 fixup(src.offsetInAtom, c, k, addend), atom(src.atom) { src.atom->incrementFixupCount(); }
1037
1038 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k) :
1039 fixup(src.offsetInAtom, c, k, (uint64_t)0), atom(src.atom) { src.atom->incrementFixupCount(); }
1040
1041 ld::Fixup fixup;
1042 Atom<A>* atom;
1043 };
1044
1045 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, Atom<A>* target) {
1046 _allFixups.push_back(FixupInAtom(src, c, k, target));
1047 }
1048
1049 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, ld::Fixup::TargetBinding b, Atom<A>* target) {
1050 _allFixups.push_back(FixupInAtom(src, c, k, b, target));
1051 }
1052
1053 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, bool wi, const char* name) {
1054 _allFixups.push_back(FixupInAtom(src, c, k, wi, name));
1055 }
1056
1057 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, ld::Fixup::TargetBinding b, const char* name) {
1058 _allFixups.push_back(FixupInAtom(src, c, k, b, name));
1059 }
1060
1061 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, uint64_t addend) {
1062 _allFixups.push_back(FixupInAtom(src, c, k, addend));
1063 }
1064
1065 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k) {
1066 _allFixups.push_back(FixupInAtom(src, c, k));
1067 }
1068
1069 const char* path() { return _path; }
1070 uint32_t symbolCount() { return _symbolCount; }
1071 uint32_t indirectSymbol(uint32_t indirectIndex);
1072 const macho_nlist<P>& symbolFromIndex(uint32_t index);
1073 const char* nameFromSymbol(const macho_nlist<P>& sym);
1074 ld::Atom::Scope scopeFromSymbol(const macho_nlist<P>& sym);
1075 static ld::Atom::Definition definitionFromSymbol(const macho_nlist<P>& sym);
1076 static ld::Atom::Combine combineFromSymbol(const macho_nlist<P>& sym);
1077 ld::Atom::SymbolTableInclusion inclusionFromSymbol(const macho_nlist<P>& sym);
1078 static bool dontDeadStripFromSymbol(const macho_nlist<P>& sym);
1079 static bool isThumbFromSymbol(const macho_nlist<P>& sym);
1080 static bool weakImportFromSymbol(const macho_nlist<P>& sym);
1081 static bool resolverFromSymbol(const macho_nlist<P>& sym);
1082 static bool altEntryFromSymbol(const macho_nlist<P>& sym);
1083 uint32_t symbolIndexFromIndirectSectionAddress(pint_t,const macho_section<P>*);
1084 const macho_section<P>* firstMachOSection() { return _sectionsStart; }
1085 const macho_section<P>* machOSectionFromSectionIndex(uint32_t index);
1086 uint32_t machOSectionCount() { return _machOSectionsCount; }
1087 uint32_t undefinedStartIndex() { return _undefinedStartIndex; }
1088 uint32_t undefinedEndIndex() { return _undefinedEndIndex; }
1089 void addFixup(FixupInAtom f) { _allFixups.push_back(f); }
1090 Section<A>* sectionForNum(unsigned int sectNum);
1091 Section<A>* sectionForAddress(pint_t addr);
1092 Atom<A>* findAtomByAddress(pint_t addr);
1093 Atom<A>* findAtomByAddressOrNullIfStub(pint_t addr);
1094 Atom<A>* findAtomByAddressOrLocalTargetOfStub(pint_t addr, uint32_t* offsetInAtom);
1095 Atom<A>* findAtomByName(const char* name); // slow!
1096 void findTargetFromAddress(pint_t addr, TargetDesc& target);
1097 void findTargetFromAddress(pint_t baseAddr, pint_t addr, TargetDesc& target);
1098 void findTargetFromAddressAndSectionNum(pint_t addr, unsigned int sectNum,
1099 TargetDesc& target);
1100 uint32_t tentativeDefinitionCount() { return _tentativeDefinitionCount; }
1101 uint32_t absoluteSymbolCount() { return _absoluteSymbolCount; }
1102
1103 uint32_t fileLength() const { return _fileLength; }
1104 bool hasStubsSection() { return (_stubsSectionNum != 0); }
1105 unsigned int stubsSectionNum() { return _stubsSectionNum; }
1106 void addDtraceExtraInfos(const SourceLocation& src, const char* provider);
1107 const char* scanSymbolTableForAddress(uint64_t addr);
1108 bool warnUnwindConversionProblems() { return _warnUnwindConversionProblems; }
1109 bool hasDataInCodeLabels() { return _hasDataInCodeLabels; }
1110 bool keepDwarfUnwind() { return _keepDwarfUnwind; }
1111 bool forceDwarfConversion() { return _forceDwarfConversion; }
1112 bool verboseOptimizationHints() { return _verboseOptimizationHints; }
1113 bool neverConvertDwarf() { return _neverConvertDwarf; }
1114 bool armUsesZeroCostExceptions() { return _armUsesZeroCostExceptions; }
1115 uint8_t maxDefaultCommonAlignment() { return _maxDefaultCommonAlignment; }
1116
1117
1118 macho_data_in_code_entry<P>* dataInCodeStart() { return _dataInCodeStart; }
1119 macho_data_in_code_entry<P>* dataInCodeEnd() { return _dataInCodeEnd; }
1120 const uint8_t* optimizationHintsStart() { return _lohStart; }
1121 const uint8_t* optimizationHintsEnd() { return _lohEnd; }
1122 bool hasOptimizationHints() { return _lohStart != _lohEnd; }
1123
1124
1125 void addFixups(const SourceLocation& src, ld::Fixup::Kind kind, const TargetDesc& target);
1126 void addFixups(const SourceLocation& src, ld::Fixup::Kind kind, const TargetDesc& target, const TargetDesc& picBase);
1127
1128
1129
1130 struct LabelAndCFIBreakIterator {
1131 typedef typename CFISection<A>::CFI_Atom_Info CFI_Atom_Info;
1132 LabelAndCFIBreakIterator(const uint32_t* ssa, uint32_t ssc, const pint_t* cfisa,
1133 uint32_t cfisc, bool ols)
1134 : sortedSymbolIndexes(ssa), sortedSymbolCount(ssc), cfiStartsArray(cfisa),
1135 cfiStartsCount(cfisc), fileHasOverlappingSymbols(ols),
1136 newSection(false), cfiIndex(0), symIndex(0) {}
1137 bool next(Parser<A>& parser, const Section<A>& sect, uint32_t sectNum, pint_t startAddr, pint_t endAddr,
1138 pint_t* addr, pint_t* size, const macho_nlist<P>** sym);
1139 pint_t peek(Parser<A>& parser, pint_t startAddr, pint_t endAddr);
1140 void beginSection() { newSection = true; symIndex = 0; }
1141
1142 const uint32_t* const sortedSymbolIndexes;
1143 const uint32_t sortedSymbolCount;
1144 const pint_t* cfiStartsArray;
1145 const uint32_t cfiStartsCount;
1146 const bool fileHasOverlappingSymbols;
1147 bool newSection;
1148 uint32_t cfiIndex;
1149 uint32_t symIndex;
1150 };
1151
1152 struct CFI_CU_InfoArrays {
1153 typedef typename CFISection<A>::CFI_Atom_Info CFI_Atom_Info;
1154 typedef typename CUSection<A>::Info CU_Info;
1155 CFI_CU_InfoArrays(const CFI_Atom_Info* cfiAr, uint32_t cfiC, CU_Info* cuAr, uint32_t cuC)
1156 : cfiArray(cfiAr), cuArray(cuAr), cfiCount(cfiC), cuCount(cuC) {}
1157 const CFI_Atom_Info* const cfiArray;
1158 CU_Info* const cuArray;
1159 const uint32_t cfiCount;
1160 const uint32_t cuCount;
1161 };
1162
1163
1164
1165 private:
1166 friend class Section<A>;
1167
1168 enum SectionType { sectionTypeIgnore, sectionTypeLiteral4, sectionTypeLiteral8, sectionTypeLiteral16,
1169 sectionTypeNonLazy, sectionTypeCFI, sectionTypeCString, sectionTypeCStringPointer,
1170 sectionTypeUTF16Strings, sectionTypeCFString, sectionTypeObjC2ClassRefs, typeObjC2CategoryList,
1171 sectionTypeObjC1Classes, sectionTypeSymboled, sectionTypeObjC1ClassRefs,
1172 sectionTypeTentativeDefinitions, sectionTypeAbsoluteSymbols, sectionTypeTLVDefs,
1173 sectionTypeCompactUnwind, sectionTypeTLVPointers};
1174
1175 template <typename P>
1176 struct MachOSectionAndSectionClass
1177 {
1178 const macho_section<P>* sect;
1179 SectionType type;
1180
1181 static int sorter(const void* l, const void* r) {
1182 const MachOSectionAndSectionClass<P>* left = (MachOSectionAndSectionClass<P>*)l;
1183 const MachOSectionAndSectionClass<P>* right = (MachOSectionAndSectionClass<P>*)r;
1184 int64_t diff = left->sect->addr() - right->sect->addr();
1185 if ( diff == 0 )
1186 return 0;
1187 if ( diff < 0 )
1188 return -1;
1189 else
1190 return 1;
1191 }
1192 };
1193
1194 struct ParserAndSectionsArray { Parser* parser; const uint32_t* sortedSectionsArray; };
1195
1196
1197 Parser(const uint8_t* fileContent, uint64_t fileLength,
1198 const char* path, time_t modTime, ld::File::Ordinal ordinal,
1199 bool warnUnwindConversionProblems, bool keepDwarfUnwind,
1200 bool forceDwarfConversion, bool neverConvertDwarf,
1201 bool verboseOptimizationHints, bool ignoreMismatchPlatform);
1202 ld::relocatable::File* parse(const ParserOptions& opts);
1203 static uint8_t loadCommandSizeMask();
1204 bool parseLoadCommands(Options::Platform platform, uint32_t minOSVersion, bool simulator, bool ignoreMismatchPlatform);
1205 void makeSections();
1206 void prescanSymbolTable();
1207 void makeSortedSymbolsArray(uint32_t symArray[], const uint32_t sectionArray[]);
1208 void makeSortedSectionsArray(uint32_t array[]);
1209 static int pointerSorter(const void* l, const void* r);
1210 static int symbolIndexSorter(void* extra, const void* l, const void* r);
1211 static int sectionIndexSorter(void* extra, const void* l, const void* r);
1212
1213 void parseDebugInfo();
1214 void parseStabs();
1215 void appendAliasAtoms(uint8_t* atomBuffer);
1216 static bool isConstFunStabs(const char *stabStr);
1217 bool read_comp_unit(const char ** name, const char ** comp_dir,
1218 uint64_t *stmt_list);
1219 pint_t realAddr(pint_t addr);
1220 const char* getDwarfString(uint64_t form, const uint8_t*& p);
1221 uint64_t getDwarfOffset(uint64_t form, const uint8_t*& di, bool dwarf64);
1222 bool skip_form(const uint8_t ** offset, const uint8_t * end,
1223 uint64_t form, uint8_t addr_size, bool dwarf64);
1224
1225
1226 // filled in by constructor
1227 const uint8_t* _fileContent;
1228 uint32_t _fileLength;
1229 const char* _path;
1230 time_t _modTime;
1231 ld::File::Ordinal _ordinal;
1232
1233 // filled in by parseLoadCommands()
1234 File<A>* _file;
1235 const macho_nlist<P>* _symbols;
1236 uint32_t _symbolCount;
1237 uint32_t _indirectSymbolCount;
1238 const char* _strings;
1239 uint32_t _stringsSize;
1240 const uint32_t* _indirectTable;
1241 uint32_t _indirectTableCount;
1242 uint32_t _undefinedStartIndex;
1243 uint32_t _undefinedEndIndex;
1244 const macho_section<P>* _sectionsStart;
1245 uint32_t _machOSectionsCount;
1246 bool _hasUUID;
1247 macho_data_in_code_entry<P>* _dataInCodeStart;
1248 macho_data_in_code_entry<P>* _dataInCodeEnd;
1249 const uint8_t* _lohStart;
1250 const uint8_t* _lohEnd;
1251
1252 // filled in by parse()
1253 CFISection<A>* _EHFrameSection;
1254 CUSection<A>* _compactUnwindSection;
1255 AbsoluteSymbolSection<A>* _absoluteSection;
1256 uint32_t _tentativeDefinitionCount;
1257 uint32_t _absoluteSymbolCount;
1258 uint32_t _symbolsInSections;
1259 bool _hasLongBranchStubs;
1260 bool _AppleObjc; // FSF has objc that uses different data layout
1261 bool _overlappingSymbols;
1262 bool _warnUnwindConversionProblems;
1263 bool _hasDataInCodeLabels;
1264 bool _keepDwarfUnwind;
1265 bool _forceDwarfConversion;
1266 bool _neverConvertDwarf;
1267 bool _verboseOptimizationHints;
1268 bool _armUsesZeroCostExceptions;
1269 bool _ignoreMismatchPlatform;
1270 bool _treateBitcodeAsData;
1271 bool _usingBitcode;
1272 uint8_t _maxDefaultCommonAlignment;
1273 unsigned int _stubsSectionNum;
1274 const macho_section<P>* _stubsMachOSection;
1275 std::vector<const char*> _dtraceProviderInfo;
1276 std::vector<FixupInAtom> _allFixups;
1277 };
1278
1279
1280
1281 template <typename A>
1282 Parser<A>::Parser(const uint8_t* fileContent, uint64_t fileLength, const char* path, time_t modTime,
1283 ld::File::Ordinal ordinal, bool convertDUI, bool keepDwarfUnwind, bool forceDwarfConversion,
1284 bool neverConvertDwarf, bool verboseOptimizationHints, bool ignoreMismatchPlatform)
1285 : _fileContent(fileContent), _fileLength(fileLength), _path(path), _modTime(modTime),
1286 _ordinal(ordinal), _file(NULL),
1287 _symbols(NULL), _symbolCount(0), _indirectSymbolCount(0), _strings(NULL), _stringsSize(0),
1288 _indirectTable(NULL), _indirectTableCount(0),
1289 _undefinedStartIndex(0), _undefinedEndIndex(0),
1290 _sectionsStart(NULL), _machOSectionsCount(0), _hasUUID(false),
1291 _dataInCodeStart(NULL), _dataInCodeEnd(NULL),
1292 _lohStart(NULL), _lohEnd(NULL),
1293 _EHFrameSection(NULL), _compactUnwindSection(NULL), _absoluteSection(NULL),
1294 _tentativeDefinitionCount(0), _absoluteSymbolCount(0),
1295 _symbolsInSections(0), _hasLongBranchStubs(false), _AppleObjc(false),
1296 _overlappingSymbols(false), _warnUnwindConversionProblems(convertDUI), _hasDataInCodeLabels(false),
1297 _keepDwarfUnwind(keepDwarfUnwind), _forceDwarfConversion(forceDwarfConversion),
1298 _neverConvertDwarf(neverConvertDwarf),
1299 _verboseOptimizationHints(verboseOptimizationHints),
1300 _ignoreMismatchPlatform(ignoreMismatchPlatform),
1301 _stubsSectionNum(0), _stubsMachOSection(NULL)
1302 {
1303 }
1304
1305
1306 template <>
1307 bool Parser<x86>::validFile(const uint8_t* fileContent, bool, cpu_subtype_t)
1308 {
1309 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1310 if ( header->magic() != MH_MAGIC )
1311 return false;
1312 if ( header->cputype() != CPU_TYPE_I386 )
1313 return false;
1314 if ( header->filetype() != MH_OBJECT )
1315 return false;
1316 return true;
1317 }
1318
1319 template <>
1320 bool Parser<x86_64>::validFile(const uint8_t* fileContent, bool, cpu_subtype_t)
1321 {
1322 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1323 if ( header->magic() != MH_MAGIC_64 )
1324 return false;
1325 if ( header->cputype() != CPU_TYPE_X86_64 )
1326 return false;
1327 if ( header->filetype() != MH_OBJECT )
1328 return false;
1329 return true;
1330 }
1331
1332 template <>
1333 bool Parser<arm>::validFile(const uint8_t* fileContent, bool subtypeMustMatch, cpu_subtype_t subtype)
1334 {
1335 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1336 if ( header->magic() != MH_MAGIC )
1337 return false;
1338 if ( header->cputype() != CPU_TYPE_ARM )
1339 return false;
1340 if ( header->filetype() != MH_OBJECT )
1341 return false;
1342 if ( subtypeMustMatch ) {
1343 if ( (cpu_subtype_t)header->cpusubtype() == subtype )
1344 return true;
1345 // hack until libcc_kext.a is made fat
1346 if ( header->cpusubtype() == CPU_SUBTYPE_ARM_ALL )
1347 return true;
1348 return false;
1349 }
1350 return true;
1351 }
1352
1353
1354 template <>
1355 bool Parser<arm64>::validFile(const uint8_t* fileContent, bool subtypeMustMatch, cpu_subtype_t subtype)
1356 {
1357 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1358 if ( header->magic() != MH_MAGIC_64 )
1359 return false;
1360 if ( header->cputype() != CPU_TYPE_ARM64 )
1361 return false;
1362 if ( header->filetype() != MH_OBJECT )
1363 return false;
1364 return true;
1365 }
1366
1367
1368 template <>
1369 const char* Parser<x86>::fileKind(const uint8_t* fileContent)
1370 {
1371 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1372 if ( header->magic() != MH_MAGIC )
1373 return NULL;
1374 if ( header->cputype() != CPU_TYPE_I386 )
1375 return NULL;
1376 return "i386";
1377 }
1378
1379 template <>
1380 const char* Parser<x86_64>::fileKind(const uint8_t* fileContent)
1381 {
1382 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1383 if ( header->magic() != MH_MAGIC_64 )
1384 return NULL;
1385 if ( header->cputype() != CPU_TYPE_X86_64 )
1386 return NULL;
1387 return "x86_64";
1388 }
1389
1390 template <>
1391 const char* Parser<arm>::fileKind(const uint8_t* fileContent)
1392 {
1393 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1394 if ( header->magic() != MH_MAGIC )
1395 return NULL;
1396 if ( header->cputype() != CPU_TYPE_ARM )
1397 return NULL;
1398 for (const ArchInfo* t=archInfoArray; t->archName != NULL; ++t) {
1399 if ( (t->cpuType == CPU_TYPE_ARM) && ((cpu_subtype_t)header->cpusubtype() == t->cpuSubType) ) {
1400 return t->archName;
1401 }
1402 }
1403 return "arm???";
1404 }
1405
1406 #if SUPPORT_ARCH_arm64
1407 template <>
1408 const char* Parser<arm64>::fileKind(const uint8_t* fileContent)
1409 {
1410 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1411 if ( header->magic() != MH_MAGIC_64 )
1412 return NULL;
1413 if ( header->cputype() != CPU_TYPE_ARM64 )
1414 return NULL;
1415 return "arm64";
1416 }
1417 #endif
1418
1419
1420 template <typename A>
1421 bool Parser<A>::hasObjC2Categories(const uint8_t* fileContent)
1422 {
1423 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1424 const uint32_t cmd_count = header->ncmds();
1425 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
1426 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
1427 const macho_load_command<P>* cmd = cmds;
1428 for (uint32_t i = 0; i < cmd_count; ++i) {
1429 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
1430 const macho_segment_command<P>* segment = (macho_segment_command<P>*)cmd;
1431 const macho_section<P>* sectionsStart = (macho_section<P>*)((char*)segment + sizeof(macho_segment_command<P>));
1432 for (uint32_t si=0; si < segment->nsects(); ++si) {
1433 const macho_section<P>* sect = &sectionsStart[si];
1434 if ( (sect->size() > 0)
1435 && (strcmp(sect->sectname(), "__objc_catlist") == 0)
1436 && (strcmp(sect->segname(), "__DATA") == 0) ) {
1437 return true;
1438 }
1439 }
1440 }
1441 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
1442 if ( cmd > cmdsEnd )
1443 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
1444 }
1445 return false;
1446 }
1447
1448
1449 template <typename A>
1450 bool Parser<A>::hasObjC1Categories(const uint8_t* fileContent)
1451 {
1452 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1453 const uint32_t cmd_count = header->ncmds();
1454 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
1455 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
1456 const macho_load_command<P>* cmd = cmds;
1457 for (uint32_t i = 0; i < cmd_count; ++i) {
1458 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
1459 const macho_segment_command<P>* segment = (macho_segment_command<P>*)cmd;
1460 const macho_section<P>* sectionsStart = (macho_section<P>*)((char*)segment + sizeof(macho_segment_command<P>));
1461 for (uint32_t si=0; si < segment->nsects(); ++si) {
1462 const macho_section<P>* sect = &sectionsStart[si];
1463 if ( (sect->size() > 0)
1464 && (strcmp(sect->sectname(), "__category") == 0)
1465 && (strcmp(sect->segname(), "__OBJC") == 0) ) {
1466 return true;
1467 }
1468 }
1469 }
1470 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
1471 if ( cmd > cmdsEnd )
1472 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
1473 }
1474 return false;
1475 }
1476
1477
1478 template <typename A>
1479 bool Parser<A>::getNonLocalSymbols(const uint8_t* fileContent, std::vector<const char*> &syms)
1480 {
1481 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1482 const uint32_t cmd_count = header->ncmds();
1483 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
1484 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
1485 const macho_load_command<P>* cmd = cmds;
1486 for (uint32_t i = 0; i < cmd_count; ++i) {
1487 if ( cmd->cmd() == LC_SYMTAB ) {
1488 const macho_symtab_command<P>* symtab = (macho_symtab_command<P>*)cmd;
1489 uint32_t symbolCount = symtab->nsyms();
1490 const macho_nlist<P>* symbols = (const macho_nlist<P>*)(fileContent + symtab->symoff());
1491 const char* strings = (char*)fileContent + symtab->stroff();
1492 for (uint32_t j = 0; j < symbolCount; ++j) {
1493 // ignore stabs and count only ext symbols
1494 if ( (symbols[j].n_type() & N_STAB) == 0 &&
1495 (symbols[j].n_type() & N_EXT) != 0 ) {
1496 const char* symName = &strings[symbols[j].n_strx()];
1497 syms.push_back(symName);
1498 }
1499 }
1500 return true;
1501 }
1502 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
1503 if ( cmd > cmdsEnd )
1504 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
1505 }
1506 return false;
1507 }
1508
1509
1510 template <typename A>
1511 int Parser<A>::pointerSorter(const void* l, const void* r)
1512 {
1513 // sort references by address
1514 const pint_t* left = (pint_t*)l;
1515 const pint_t* right = (pint_t*)r;
1516 return (*left - *right);
1517 }
1518
1519 template <typename A>
1520 typename A::P::uint_t Parser<A>::LabelAndCFIBreakIterator::peek(Parser<A>& parser, pint_t startAddr, pint_t endAddr)
1521 {
1522 pint_t symbolAddr;
1523 if ( symIndex < sortedSymbolCount )
1524 symbolAddr = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]).n_value();
1525 else
1526 symbolAddr = endAddr;
1527 pint_t cfiAddr;
1528 if ( cfiIndex < cfiStartsCount )
1529 cfiAddr = cfiStartsArray[cfiIndex];
1530 else
1531 cfiAddr = endAddr;
1532 if ( (cfiAddr < symbolAddr) && (cfiAddr >= startAddr) ) {
1533 if ( cfiAddr < endAddr )
1534 return cfiAddr;
1535 else
1536 return endAddr;
1537 }
1538 else {
1539 if ( symbolAddr < endAddr )
1540 return symbolAddr;
1541 else
1542 return endAddr;
1543 }
1544 }
1545
1546 //
1547 // Parses up a section into chunks based on labels and CFI information.
1548 // Each call returns the next chunk address and size, and (if the break
1549 // was becuase of a label, the symbol). Returns false when no more chunks.
1550 //
1551 template <typename A>
1552 bool Parser<A>::LabelAndCFIBreakIterator::next(Parser<A>& parser, const Section<A>& sect, uint32_t sectNum, pint_t startAddr, pint_t endAddr,
1553 pint_t* addr, pint_t* size, const macho_nlist<P>** symbol)
1554 {
1555 // may not be a label on start of section, but need atom demarcation there
1556 if ( newSection ) {
1557 newSection = false;
1558 // advance symIndex until we get to the first label at or past the start of this section
1559 while ( symIndex < sortedSymbolCount ) {
1560 const macho_nlist<P>& sym = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]);
1561 if ( ! sect.ignoreLabel(parser.nameFromSymbol(sym)) ) {
1562 pint_t nextSymbolAddr = sym.n_value();
1563 //fprintf(stderr, "sectNum=%d, nextSymbolAddr=0x%08llX, name=%s\n", sectNum, (uint64_t)nextSymbolAddr, parser.nameFromSymbol(sym));
1564 if ( (nextSymbolAddr > startAddr) || ((nextSymbolAddr == startAddr) && (sym.n_sect() == sectNum)) )
1565 break;
1566 }
1567 ++symIndex;
1568 }
1569 if ( symIndex < sortedSymbolCount ) {
1570 const macho_nlist<P>& sym = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]);
1571 pint_t nextSymbolAddr = sym.n_value();
1572 // if next symbol found is not in this section
1573 if ( sym.n_sect() != sectNum ) {
1574 // check for CFI break instead of symbol break
1575 if ( cfiIndex < cfiStartsCount ) {
1576 pint_t nextCfiAddr = cfiStartsArray[cfiIndex];
1577 if ( nextCfiAddr < endAddr ) {
1578 // use cfi
1579 ++cfiIndex;
1580 *addr = nextCfiAddr;
1581 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1582 *symbol = NULL;
1583 return true;
1584 }
1585 }
1586 *addr = startAddr;
1587 *size = endAddr - startAddr;
1588 *symbol = NULL;
1589 if ( startAddr == endAddr )
1590 return false; // zero size section
1591 else
1592 return true; // whole section is one atom with no label
1593 }
1594 // if also CFI break here, eat it
1595 if ( cfiIndex < cfiStartsCount ) {
1596 if ( cfiStartsArray[cfiIndex] == nextSymbolAddr )
1597 ++cfiIndex;
1598 }
1599 if ( nextSymbolAddr == startAddr ) {
1600 // label at start of section, return it as chunk
1601 ++symIndex;
1602 *addr = startAddr;
1603 *size = peek(parser, startAddr, endAddr) - startAddr;
1604 *symbol = &sym;
1605 return true;
1606 }
1607 // return chunk before first symbol
1608 *addr = startAddr;
1609 *size = nextSymbolAddr - startAddr;
1610 *symbol = NULL;
1611 return true;
1612 }
1613 // no symbols in section, check CFI
1614 if ( cfiIndex < cfiStartsCount ) {
1615 pint_t nextCfiAddr = cfiStartsArray[cfiIndex];
1616 if ( nextCfiAddr < endAddr ) {
1617 // use cfi
1618 ++cfiIndex;
1619 *addr = nextCfiAddr;
1620 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1621 *symbol = NULL;
1622 return true;
1623 }
1624 }
1625 // no cfi, so whole section is one chunk
1626 *addr = startAddr;
1627 *size = endAddr - startAddr;
1628 *symbol = NULL;
1629 if ( startAddr == endAddr )
1630 return false; // zero size section
1631 else
1632 return true; // whole section is one atom with no label
1633 }
1634
1635 while ( (symIndex < sortedSymbolCount) && (cfiIndex < cfiStartsCount) ) {
1636 const macho_nlist<P>& sym = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]);
1637 pint_t nextSymbolAddr = sym.n_value();
1638 pint_t nextCfiAddr = cfiStartsArray[cfiIndex];
1639 if ( nextSymbolAddr < nextCfiAddr ) {
1640 if ( nextSymbolAddr >= endAddr )
1641 return false;
1642 ++symIndex;
1643 if ( nextSymbolAddr < startAddr )
1644 continue;
1645 *addr = nextSymbolAddr;
1646 *size = peek(parser, startAddr, endAddr) - nextSymbolAddr;
1647 *symbol = &sym;
1648 return true;
1649 }
1650 else if ( nextCfiAddr < nextSymbolAddr ) {
1651 if ( nextCfiAddr >= endAddr )
1652 return false;
1653 ++cfiIndex;
1654 if ( nextCfiAddr < startAddr )
1655 continue;
1656 *addr = nextCfiAddr;
1657 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1658 *symbol = NULL;
1659 return true;
1660 }
1661 else {
1662 if ( nextCfiAddr >= endAddr )
1663 return false;
1664 ++symIndex;
1665 ++cfiIndex;
1666 if ( nextCfiAddr < startAddr )
1667 continue;
1668 *addr = nextCfiAddr;
1669 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1670 *symbol = &sym;
1671 return true;
1672 }
1673 }
1674 while ( symIndex < sortedSymbolCount ) {
1675 const macho_nlist<P>& sym = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]);
1676 pint_t nextSymbolAddr = sym.n_value();
1677 // if next symbol found is not in this section, then done with iteration
1678 if ( sym.n_sect() != sectNum )
1679 return false;
1680 ++symIndex;
1681 if ( nextSymbolAddr < startAddr )
1682 continue;
1683 *addr = nextSymbolAddr;
1684 *size = peek(parser, startAddr, endAddr) - nextSymbolAddr;
1685 *symbol = &sym;
1686 return true;
1687 }
1688 while ( cfiIndex < cfiStartsCount ) {
1689 pint_t nextCfiAddr = cfiStartsArray[cfiIndex];
1690 if ( nextCfiAddr >= endAddr )
1691 return false;
1692 ++cfiIndex;
1693 if ( nextCfiAddr < startAddr )
1694 continue;
1695 *addr = nextCfiAddr;
1696 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1697 *symbol = NULL;
1698 return true;
1699 }
1700 return false;
1701 }
1702
1703 template <>
1704 typename arm::P::uint_t Parser<arm>::realAddr(typename arm::P::uint_t addr)
1705 {
1706 return addr & (-2);
1707 }
1708
1709 template <typename A>
1710 typename A::P::uint_t Parser<A>::realAddr(typename A::P::uint_t addr)
1711 {
1712 return addr;
1713 }
1714
1715 #define STACK_ALLOC_IF_SMALL(_type, _name, _actual_count, _maxCount) \
1716 _type* _name = NULL; \
1717 uint32_t _name##_count = 1; \
1718 if ( _actual_count > _maxCount ) \
1719 _name = (_type*)malloc(sizeof(_type) * _actual_count); \
1720 else \
1721 _name##_count = _actual_count; \
1722 _type _name##_buffer[_name##_count]; \
1723 if ( _name == NULL ) \
1724 _name = _name##_buffer;
1725
1726
1727 template <typename A>
1728 ld::relocatable::File* Parser<A>::parse(const ParserOptions& opts)
1729 {
1730 // create file object
1731 _file = new File<A>(_path, _modTime, _fileContent, _ordinal);
1732
1733 // set sourceKind
1734 _file->_srcKind = opts.srcKind;
1735 // set treatBitcodeAsData
1736 _treateBitcodeAsData = opts.treateBitcodeAsData;
1737 _usingBitcode = opts.usingBitcode;
1738
1739 // respond to -t option
1740 if ( opts.logAllFiles )
1741 printf("%s\n", _path);
1742
1743 _armUsesZeroCostExceptions = opts.armUsesZeroCostExceptions;
1744 _maxDefaultCommonAlignment = opts.maxDefaultCommonAlignment;
1745
1746 // parse start of mach-o file
1747 if ( ! parseLoadCommands(opts.platform, opts.minOSVersion, opts.simulator, opts.ignoreMismatchPlatform) )
1748 return _file;
1749
1750 // make array of
1751 uint32_t sortedSectionIndexes[_machOSectionsCount];
1752 this->makeSortedSectionsArray(sortedSectionIndexes);
1753
1754 // make symbol table sorted by address
1755 this->prescanSymbolTable();
1756 uint32_t sortedSymbolIndexes[_symbolsInSections];
1757 this->makeSortedSymbolsArray(sortedSymbolIndexes, sortedSectionIndexes);
1758
1759 // allocate Section<A> object for each mach-o section
1760 makeSections();
1761
1762 // if it exists, do special early parsing of __compact_unwind section
1763 uint32_t countOfCUs = 0;
1764 if ( _compactUnwindSection != NULL )
1765 countOfCUs = _compactUnwindSection->count();
1766 // stack allocate (if not too large) cuInfoBuffer
1767 STACK_ALLOC_IF_SMALL(typename CUSection<A>::Info, cuInfoArray, countOfCUs, 1024);
1768 if ( countOfCUs != 0 )
1769 _compactUnwindSection->parse(*this, countOfCUs, cuInfoArray);
1770
1771 // create lists of address that already have compact unwind and thus don't need the dwarf parsed
1772 unsigned cuLsdaCount = 0;
1773 pint_t cuStarts[countOfCUs];
1774 for (uint32_t i=0; i < countOfCUs; ++i) {
1775 if ( CUSection<A>::encodingMeansUseDwarf(cuInfoArray[i].compactUnwindInfo) )
1776 cuStarts[i] = -1;
1777 else
1778 cuStarts[i] = cuInfoArray[i].functionStartAddress;
1779 if ( cuInfoArray[i].lsdaAddress != 0 )
1780 ++cuLsdaCount;
1781 }
1782
1783
1784 // if it exists, do special early parsing of __eh_frame section
1785 // stack allocate (if not too large) array of CFI_Atom_Info
1786 uint32_t countOfCFIs = 0;
1787 if ( _EHFrameSection != NULL )
1788 countOfCFIs = _EHFrameSection->cfiCount(*this);
1789 STACK_ALLOC_IF_SMALL(typename CFISection<A>::CFI_Atom_Info, cfiArray, countOfCFIs, 1024);
1790
1791 // stack allocate (if not too large) a copy of __eh_frame to apply relocations to
1792 uint32_t sectSize = 4;
1793 if ( (countOfCFIs != 0) && _EHFrameSection->needsRelocating() )
1794 sectSize = _EHFrameSection->machoSection()->size()+4;
1795 STACK_ALLOC_IF_SMALL(uint8_t, ehBuffer, sectSize, 50*1024);
1796 uint32_t cfiStartsCount = 0;
1797 if ( countOfCFIs != 0 ) {
1798 _EHFrameSection->cfiParse(*this, ehBuffer, cfiArray, countOfCFIs, cuStarts, countOfCUs);
1799 // count functions and lsdas
1800 for(uint32_t i=0; i < countOfCFIs; ++i) {
1801 if ( cfiArray[i].isCIE )
1802 continue;
1803 //fprintf(stderr, "cfiArray[i].func = 0x%08llX, cfiArray[i].lsda = 0x%08llX, encoding=0x%08X\n",
1804 // (uint64_t)cfiArray[i].u.fdeInfo.function.targetAddress,
1805 // (uint64_t)cfiArray[i].u.fdeInfo.lsda.targetAddress,
1806 // cfiArray[i].u.fdeInfo.compactUnwindInfo);
1807 if ( cfiArray[i].u.fdeInfo.function.targetAddress != CFI_INVALID_ADDRESS )
1808 ++cfiStartsCount;
1809 if ( cfiArray[i].u.fdeInfo.lsda.targetAddress != CFI_INVALID_ADDRESS )
1810 ++cfiStartsCount;
1811 }
1812 }
1813 CFI_CU_InfoArrays cfis(cfiArray, countOfCFIs, cuInfoArray, countOfCUs);
1814
1815 // create sorted array of function starts and lsda starts
1816 pint_t cfiStartsArray[cfiStartsCount+cuLsdaCount];
1817 uint32_t countOfFDEs = 0;
1818 uint32_t cfiStartsArrayCount = 0;
1819 if ( countOfCFIs != 0 ) {
1820 for(uint32_t i=0; i < countOfCFIs; ++i) {
1821 if ( cfiArray[i].isCIE )
1822 continue;
1823 if ( cfiArray[i].u.fdeInfo.function.targetAddress != CFI_INVALID_ADDRESS )
1824 cfiStartsArray[cfiStartsArrayCount++] = realAddr(cfiArray[i].u.fdeInfo.function.targetAddress);
1825 if ( cfiArray[i].u.fdeInfo.lsda.targetAddress != CFI_INVALID_ADDRESS )
1826 cfiStartsArray[cfiStartsArrayCount++] = cfiArray[i].u.fdeInfo.lsda.targetAddress;
1827 ++countOfFDEs;
1828 }
1829 }
1830 if ( cuLsdaCount != 0 ) {
1831 // merge in an lsda info from compact unwind
1832 for (uint32_t i=0; i < countOfCUs; ++i) {
1833 if ( cuInfoArray[i].lsdaAddress == 0 )
1834 continue;
1835 // append to cfiStartsArray if not already in that list
1836 bool found = false;
1837 for(uint32_t j=0; j < cfiStartsArrayCount; ++j) {
1838 if ( cfiStartsArray[j] == cuInfoArray[i].lsdaAddress )
1839 found = true;
1840 }
1841 if ( ! found ) {
1842 cfiStartsArray[cfiStartsArrayCount++] = cuInfoArray[i].lsdaAddress;
1843 }
1844 }
1845 }
1846 if ( cfiStartsArrayCount != 0 ) {
1847 ::qsort(cfiStartsArray, cfiStartsArrayCount, sizeof(pint_t), pointerSorter);
1848 #ifndef NDEBUG
1849 // scan for FDEs claming the same function
1850 for(uint32_t i=1; i < cfiStartsArrayCount; ++i) {
1851 assert( cfiStartsArray[i] != cfiStartsArray[i-1] );
1852 }
1853 #endif
1854 }
1855
1856 Section<A>** sections = _file->_sectionsArray;
1857 uint32_t sectionsCount = _file->_sectionsArrayCount;
1858
1859 // figure out how many atoms will be allocated and allocate
1860 LabelAndCFIBreakIterator breakIterator(sortedSymbolIndexes, _symbolsInSections, cfiStartsArray,
1861 cfiStartsArrayCount, _overlappingSymbols);
1862 uint32_t computedAtomCount = 0;
1863 for (uint32_t i=0; i < sectionsCount; ++i ) {
1864 breakIterator.beginSection();
1865 uint32_t count = sections[i]->computeAtomCount(*this, breakIterator, cfis);
1866 //const macho_section<P>* sect = sections[i]->machoSection();
1867 //fprintf(stderr, "computed count=%u for section %s size=%llu\n", count, sect->sectname(), (sect != NULL) ? sect->size() : 0);
1868 computedAtomCount += count;
1869 }
1870 //fprintf(stderr, "allocating %d atoms * sizeof(Atom<A>)=%ld, sizeof(ld::Atom)=%ld\n", computedAtomCount, sizeof(Atom<A>), sizeof(ld::Atom));
1871 _file->_atomsArray = new uint8_t[computedAtomCount*sizeof(Atom<A>)];
1872 _file->_atomsArrayCount = 0;
1873
1874 // have each section append atoms to _atomsArray
1875 LabelAndCFIBreakIterator breakIterator2(sortedSymbolIndexes, _symbolsInSections, cfiStartsArray,
1876 cfiStartsArrayCount, _overlappingSymbols);
1877 for (uint32_t i=0; i < sectionsCount; ++i ) {
1878 uint8_t* atoms = _file->_atomsArray + _file->_atomsArrayCount*sizeof(Atom<A>);
1879 breakIterator2.beginSection();
1880 uint32_t count = sections[i]->appendAtoms(*this, atoms, breakIterator2, cfis);
1881 //fprintf(stderr, "append count=%u for section %s/%s\n", count, sections[i]->machoSection()->segname(), sections[i]->machoSection()->sectname());
1882 _file->_atomsArrayCount += count;
1883 }
1884 assert( _file->_atomsArrayCount == computedAtomCount && "more atoms allocated than expected");
1885
1886
1887 // have each section add all fix-ups for its atoms
1888 _allFixups.reserve(computedAtomCount*5);
1889 for (uint32_t i=0; i < sectionsCount; ++i )
1890 sections[i]->makeFixups(*this, cfis);
1891
1892 // assign fixups start offset for each atom
1893 uint8_t* p = _file->_atomsArray;
1894 uint32_t fixupOffset = 0;
1895 for(int i=_file->_atomsArrayCount; i > 0; --i) {
1896 Atom<A>* atom = (Atom<A>*)p;
1897 atom->_fixupsStartIndex = fixupOffset;
1898 fixupOffset += atom->_fixupsCount;
1899 atom->_fixupsCount = 0;
1900 p += sizeof(Atom<A>);
1901 }
1902 assert(fixupOffset == _allFixups.size());
1903 _file->_fixups.resize(fixupOffset);
1904
1905 // copy each fixup for each atom
1906 for(typename std::vector<FixupInAtom>::iterator it=_allFixups.begin(); it != _allFixups.end(); ++it) {
1907 uint32_t slot = it->atom->_fixupsStartIndex + it->atom->_fixupsCount;
1908 _file->_fixups[slot] = it->fixup;
1909 it->atom->_fixupsCount++;
1910 }
1911
1912 // done with temp vector
1913 _allFixups.clear();
1914
1915 // add unwind info
1916 _file->_unwindInfos.reserve(countOfFDEs+countOfCUs);
1917 for(uint32_t i=0; i < countOfCFIs; ++i) {
1918 if ( cfiArray[i].isCIE )
1919 continue;
1920 if ( cfiArray[i].u.fdeInfo.function.targetAddress != CFI_INVALID_ADDRESS ) {
1921 ld::Atom::UnwindInfo info;
1922 info.startOffset = 0;
1923 info.unwindInfo = cfiArray[i].u.fdeInfo.compactUnwindInfo;
1924 _file->_unwindInfos.push_back(info);
1925 Atom<A>* func = findAtomByAddress(cfiArray[i].u.fdeInfo.function.targetAddress);
1926 func->setUnwindInfoRange(_file->_unwindInfos.size()-1, 1);
1927 //fprintf(stderr, "cu from dwarf =0x%08X, atom=%s\n", info.unwindInfo, func->name());
1928 }
1929 }
1930 // apply compact infos in __LD,__compact_unwind section to each function
1931 // if function also has dwarf unwind, CU will override it
1932 Atom<A>* lastFunc = NULL;
1933 uint32_t lastEnd = 0;
1934 for(uint32_t i=0; i < countOfCUs; ++i) {
1935 typename CUSection<A>::Info* info = &cuInfoArray[i];
1936 assert(info->function != NULL);
1937 ld::Atom::UnwindInfo ui;
1938 ui.startOffset = info->functionStartAddress - info->function->objectAddress();
1939 ui.unwindInfo = info->compactUnwindInfo;
1940 _file->_unwindInfos.push_back(ui);
1941 // don't override with converted cu with "use dwarf" cu, if forcing dwarf conversion
1942 if ( !_forceDwarfConversion || !CUSection<A>::encodingMeansUseDwarf(info->compactUnwindInfo) ) {
1943 //fprintf(stderr, "cu=0x%08X, atom=%s\n", ui.unwindInfo, info->function->name());
1944 // if previous is for same function, extend range
1945 if ( info->function == lastFunc ) {
1946 if ( lastEnd != ui.startOffset ) {
1947 if ( lastEnd < ui.startOffset )
1948 warning("__LD,__compact_unwind entries for %s have a gap at offset 0x%0X", info->function->name(), lastEnd);
1949 else
1950 warning("__LD,__compact_unwind entries for %s overlap at offset 0x%0X", info->function->name(), lastEnd);
1951 }
1952 lastFunc->extendUnwindInfoRange();
1953 }
1954 else
1955 info->function->setUnwindInfoRange(_file->_unwindInfos.size()-1, 1);
1956 lastFunc = info->function;
1957 lastEnd = ui.startOffset + info->rangeLength;
1958 }
1959 }
1960
1961 // process indirect symbols which become AliasAtoms
1962 _file->_aliasAtomsArray = NULL;
1963 _file->_aliasAtomsArrayCount = 0;
1964 if ( _indirectSymbolCount != 0 ) {
1965 _file->_aliasAtomsArrayCount = _indirectSymbolCount;
1966 _file->_aliasAtomsArray = new uint8_t[_file->_aliasAtomsArrayCount*sizeof(AliasAtom)];
1967 this->appendAliasAtoms(_file->_aliasAtomsArray);
1968 }
1969
1970
1971 // parse dwarf debug info to get line info
1972 this->parseDebugInfo();
1973
1974 return _file;
1975 }
1976
1977 static void versionToString(uint32_t value, char buffer[32])
1978 {
1979 if ( value & 0xFF )
1980 sprintf(buffer, "%d.%d.%d", value >> 16, (value >> 8) & 0xFF, value & 0xFF);
1981 else
1982 sprintf(buffer, "%d.%d", value >> 16, (value >> 8) & 0xFF);
1983 }
1984
1985 template <> uint8_t Parser<x86>::loadCommandSizeMask() { return 0x03; }
1986 template <> uint8_t Parser<x86_64>::loadCommandSizeMask() { return 0x07; }
1987 template <> uint8_t Parser<arm>::loadCommandSizeMask() { return 0x03; }
1988 template <> uint8_t Parser<arm64>::loadCommandSizeMask() { return 0x07; }
1989
1990 template <typename A>
1991 bool Parser<A>::parseLoadCommands(Options::Platform platform, uint32_t linkMinOSVersion, bool simulator, bool ignoreMismatchPlatform)
1992 {
1993 const macho_header<P>* header = (const macho_header<P>*)_fileContent;
1994
1995 // set File attributes
1996 _file->_canScatterAtoms = (header->flags() & MH_SUBSECTIONS_VIA_SYMBOLS);
1997 _file->_cpuSubType = header->cpusubtype();
1998
1999 const macho_segment_command<P>* segment = NULL;
2000 const uint8_t* const endOfFile = _fileContent + _fileLength;
2001 const uint32_t cmd_count = header->ncmds();
2002 // <rdar://problem/5394172> an empty .o file with zero load commands will crash linker
2003 if ( cmd_count == 0 )
2004 return false;
2005 Options::Platform lcPlatform = Options::kPlatformUnknown;
2006 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
2007 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
2008 const macho_load_command<P>* cmd = cmds;
2009 for (uint32_t i = 0; i < cmd_count; ++i) {
2010 uint32_t size = cmd->cmdsize();
2011 if ( (size & this->loadCommandSizeMask()) != 0 )
2012 throwf("load command #%d has a unaligned size", i);
2013 const uint8_t* endOfCmd = ((uint8_t*)cmd)+cmd->cmdsize();
2014 if ( endOfCmd > (uint8_t*)cmdsEnd )
2015 throwf("load command #%d extends beyond the end of the load commands", i);
2016 if ( endOfCmd > endOfFile )
2017 throwf("load command #%d extends beyond the end of the file", i);
2018 switch (cmd->cmd()) {
2019 case LC_SYMTAB:
2020 {
2021 const macho_symtab_command<P>* symtab = (macho_symtab_command<P>*)cmd;
2022 _symbolCount = symtab->nsyms();
2023 _symbols = (const macho_nlist<P>*)(_fileContent + symtab->symoff());
2024 _strings = (char*)_fileContent + symtab->stroff();
2025 _stringsSize = symtab->strsize();
2026 if ( (symtab->symoff() + _symbolCount*sizeof(macho_nlist<P>)) > _fileLength )
2027 throw "mach-o symbol table extends beyond end of file";
2028 if ( (_strings + _stringsSize) > (char*)endOfFile )
2029 throw "mach-o string pool extends beyond end of file";
2030 if ( _indirectTable == NULL ) {
2031 if ( _undefinedEndIndex == 0 ) {
2032 _undefinedStartIndex = 0;
2033 _undefinedEndIndex = symtab->nsyms();
2034 }
2035 }
2036 }
2037 break;
2038 case LC_DYSYMTAB:
2039 {
2040 const macho_dysymtab_command<P>* dsymtab = (macho_dysymtab_command<P>*)cmd;
2041 _indirectTable = (uint32_t*)(_fileContent + dsymtab->indirectsymoff());
2042 _indirectTableCount = dsymtab->nindirectsyms();
2043 if ( &_indirectTable[_indirectTableCount] > (uint32_t*)endOfFile )
2044 throw "indirect symbol table extends beyond end of file";
2045 _undefinedStartIndex = dsymtab->iundefsym();
2046 _undefinedEndIndex = _undefinedStartIndex + dsymtab->nundefsym();
2047 }
2048 break;
2049 case LC_UUID:
2050 _hasUUID = true;
2051 break;
2052 case LC_DATA_IN_CODE:
2053 {
2054 const macho_linkedit_data_command<P>* dc = (macho_linkedit_data_command<P>*)cmd;
2055 _dataInCodeStart = (macho_data_in_code_entry<P>*)(_fileContent + dc->dataoff());
2056 _dataInCodeEnd = (macho_data_in_code_entry<P>*)(_fileContent + dc->dataoff() + dc->datasize());
2057 if ( _dataInCodeEnd > (macho_data_in_code_entry<P>*)endOfFile )
2058 throw "LC_DATA_IN_CODE table extends beyond end of file";
2059 }
2060 break;
2061 case LC_LINKER_OPTION:
2062 {
2063 const macho_linker_option_command<P>* loc = (macho_linker_option_command<P>*)cmd;
2064 const char* buffer = loc->buffer();
2065 _file->_linkerOptions.resize(_file->_linkerOptions.size() + 1);
2066 std::vector<const char*>& vec = _file->_linkerOptions.back();
2067 for (uint32_t j=0; j < loc->count(); ++j) {
2068 vec.push_back(buffer);
2069 buffer += strlen(buffer) + 1;
2070 }
2071 if ( buffer > ((char*)cmd + loc->cmdsize()) )
2072 throw "malformed LC_LINKER_OPTION";
2073 }
2074 break;
2075 case LC_LINKER_OPTIMIZATION_HINTS:
2076 {
2077 const macho_linkedit_data_command<P>* loh = (macho_linkedit_data_command<P>*)cmd;
2078 _lohStart = _fileContent + loh->dataoff();
2079 _lohEnd = _fileContent + loh->dataoff() + loh->datasize();
2080 if ( _lohEnd > endOfFile )
2081 throw "LC_LINKER_OPTIMIZATION_HINTS table extends beyond end of file";
2082 }
2083 break;
2084 case LC_VERSION_MIN_MACOSX:
2085 case LC_VERSION_MIN_IPHONEOS:
2086 case LC_VERSION_MIN_WATCHOS:
2087 #if SUPPORT_APPLE_TV
2088 case LC_VERSION_MIN_TVOS:
2089 #endif
2090 if ( ignoreMismatchPlatform )
2091 break;
2092 _file->_platform = cmd->cmd();
2093 lcPlatform = Options::platformForLoadCommand(cmd->cmd());
2094 _file->_minOSVersion = ((macho_version_min_command<P>*)cmd)->version();
2095 break;
2096 case macho_segment_command<P>::CMD:
2097 if ( segment != NULL )
2098 throw "more than one LC_SEGMENT found in object file";
2099 segment = (macho_segment_command<P>*)cmd;
2100 break;
2101 default:
2102 // ignore unknown load commands
2103 break;
2104 }
2105 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
2106 if ( cmd > cmdsEnd )
2107 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
2108 }
2109 // arm/arm64 objects are default to ios platform if not set.
2110 // rdar://problem/21746314
2111 if (lcPlatform == Options::kPlatformUnknown &&
2112 (std::is_same<A, arm>::value || std::is_same<A, arm64>::value))
2113 lcPlatform = Options::kPlatformiOS;
2114
2115 // Check platform cross-linking.
2116 if ( !ignoreMismatchPlatform ) {
2117 if ( lcPlatform != platform ) {
2118 switch (platform) {
2119 case Options::kPlatformOSX:
2120 case Options::kPlatformiOS:
2121 if ( lcPlatform == Options::kPlatformUnknown )
2122 break;
2123 // fall through if the Platform is not Unknown
2124 case Options::kPlatformWatchOS:
2125 // Error when using bitcocde, warning otherwise.
2126 if (_usingBitcode)
2127 throwf("building for %s%s, but linking in object file built for %s,",
2128 Options::platformName(platform), (simulator ? " simulator" : ""),
2129 Options::platformName(lcPlatform));
2130 else
2131 warning("URGENT: building for %s%s, but linking in object file (%s) built for %s. "
2132 "Note: This will be an error in the future.",
2133 Options::platformName(platform), (simulator ? " simulator" : ""), path(),
2134 Options::platformName(lcPlatform));
2135 break;
2136 #if SUPPORT_APPLE_TV
2137 case Options::kPlatform_tvOS:
2138 // Error when using bitcocde, warning otherwise.
2139 if (_usingBitcode)
2140 throwf("building for %s%s, but linking in object file built for %s,",
2141 Options::platformName(platform), (simulator ? " simulator" : ""),
2142 Options::platformName(lcPlatform));
2143 else
2144 warning("URGENT: building for %s%s, but linking in object file (%s) built for %s. "
2145 "Note: This will be an error in the future.",
2146 Options::platformName(platform), (simulator ? " simulator" : ""), path(),
2147 Options::platformName(lcPlatform));
2148 break;
2149 #endif
2150 case Options::kPlatformUnknown:
2151 // skip if the target platform is unknown
2152 break;
2153 }
2154 }
2155 if ( linkMinOSVersion && (_file->_minOSVersion > linkMinOSVersion) ) {
2156 char t1[32];
2157 char t2[32];
2158 versionToString(_file->_minOSVersion, t1);
2159 versionToString(linkMinOSVersion, t2);
2160 warning("object file (%s) was built for newer %s version (%s) than being linked (%s)",
2161 _path, Options::platformName(lcPlatform), t1, t2);
2162 }
2163 }
2164
2165
2166 // record range of sections
2167 if ( segment == NULL )
2168 throw "missing LC_SEGMENT";
2169 _sectionsStart = (macho_section<P>*)((char*)segment + sizeof(macho_segment_command<P>));
2170 _machOSectionsCount = segment->nsects();
2171 if ( (sizeof(macho_segment_command<P>) + _machOSectionsCount * sizeof(macho_section<P>)) > segment->cmdsize() )
2172 throw "too many sections for size of LC_SEGMENT command";
2173 return true;
2174 }
2175
2176 template <typename A>
2177 Options::Platform Parser<A>::findPlatform(const macho_header<P>* header)
2178 {
2179 const uint32_t cmd_count = header->ncmds();
2180 if ( cmd_count == 0 )
2181 return Options::kPlatformUnknown;
2182 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
2183 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
2184 const macho_load_command<P>* cmd = cmds;
2185 for (uint32_t i = 0; i < cmd_count; ++i) {
2186 uint32_t size = cmd->cmdsize();
2187 if ( (size & loadCommandSizeMask()) != 0 )
2188 throwf("load command #%d has a unaligned size", i);
2189 const uint8_t* endOfCmd = ((uint8_t*)cmd)+cmd->cmdsize();
2190 if ( endOfCmd > (uint8_t*)cmdsEnd )
2191 throwf("load command #%d extends beyond the end of the load commands", i);
2192 switch (cmd->cmd()) {
2193 case LC_VERSION_MIN_MACOSX:
2194 return Options::kPlatformOSX;
2195 case LC_VERSION_MIN_IPHONEOS:
2196 return Options::kPlatformiOS;
2197 }
2198 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
2199 if ( cmd > cmdsEnd )
2200 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
2201 }
2202 return Options::kPlatformUnknown;
2203 }
2204
2205
2206 template <typename A>
2207 void Parser<A>::prescanSymbolTable()
2208 {
2209 _tentativeDefinitionCount = 0;
2210 _absoluteSymbolCount = 0;
2211 _symbolsInSections = 0;
2212 _hasDataInCodeLabels = false;
2213 for (uint32_t i=0; i < this->_symbolCount; ++i) {
2214 const macho_nlist<P>& sym = symbolFromIndex(i);
2215 // ignore stabs
2216 if ( (sym.n_type() & N_STAB) != 0 )
2217 continue;
2218
2219 // look at undefines
2220 const char* symbolName = this->nameFromSymbol(sym);
2221 if ( (sym.n_type() & N_TYPE) == N_UNDF ) {
2222 if ( sym.n_value() != 0 ) {
2223 // count tentative definitions
2224 ++_tentativeDefinitionCount;
2225 }
2226 else if ( strncmp(symbolName, "___dtrace_", 10) == 0 ) {
2227 // any undefined starting with __dtrace_*$ that is not ___dtrace_probe$* or ___dtrace_isenabled$*
2228 // is extra provider info
2229 if ( (strncmp(&symbolName[10], "probe$", 6) != 0) && (strncmp(&symbolName[10], "isenabled$", 10) != 0) ) {
2230 _dtraceProviderInfo.push_back(symbolName);
2231 }
2232 }
2233 continue;
2234 }
2235 else if ( ((sym.n_type() & N_TYPE) == N_INDR) && ((sym.n_type() & N_EXT) != 0) ) {
2236 _indirectSymbolCount++;
2237 continue;
2238 }
2239
2240 // count absolute symbols
2241 if ( (sym.n_type() & N_TYPE) == N_ABS ) {
2242 const char* absName = this->nameFromSymbol(sym);
2243 // ignore .objc_class_name_* symbols
2244 if ( strncmp(absName, ".objc_class_name_", 17) == 0 ) {
2245 _AppleObjc = true;
2246 continue;
2247 }
2248 // ignore .objc_class_name_* symbols
2249 if ( strncmp(absName, ".objc_category_name_", 20) == 0 )
2250 continue;
2251 // ignore empty *.eh symbols
2252 if ( strcmp(&absName[strlen(absName)-3], ".eh") == 0 )
2253 continue;
2254 ++_absoluteSymbolCount;
2255 }
2256
2257 // only look at definitions
2258 if ( (sym.n_type() & N_TYPE) != N_SECT )
2259 continue;
2260
2261 // 'L' labels do not denote atom breaks
2262 if ( symbolName[0] == 'L' ) {
2263 // <rdar://problem/9218847> Formalize data in code with L$start$ labels
2264 if ( strncmp(symbolName, "L$start$", 8) == 0 )
2265 _hasDataInCodeLabels = true;
2266 continue;
2267 }
2268 // how many def syms in each section
2269 if ( sym.n_sect() > _machOSectionsCount )
2270 throw "bad n_sect in symbol table";
2271
2272 _symbolsInSections++;
2273 }
2274 }
2275
2276 template <typename A>
2277 void Parser<A>::appendAliasAtoms(uint8_t* p)
2278 {
2279 for (uint32_t i=0; i < this->_symbolCount; ++i) {
2280 const macho_nlist<P>& sym = symbolFromIndex(i);
2281 // ignore stabs
2282 if ( (sym.n_type() & N_STAB) != 0 )
2283 continue;
2284
2285 // only look at N_INDR symbols
2286 if ( (sym.n_type() & N_TYPE) != N_INDR )
2287 continue;
2288
2289 // skip non-external aliases
2290 if ( (sym.n_type() & N_EXT) == 0 )
2291 continue;
2292
2293 const char* symbolName = this->nameFromSymbol(sym);
2294 const char* aliasOfName = &_strings[sym.n_value()];
2295 bool isHiddenVisibility = (sym.n_type() & N_PEXT);
2296 AliasAtom* allocatedSpace = (AliasAtom*)p;
2297 new (allocatedSpace) AliasAtom(symbolName, isHiddenVisibility, _file, aliasOfName);
2298 p += sizeof(AliasAtom);
2299 }
2300 }
2301
2302
2303
2304 template <typename A>
2305 int Parser<A>::sectionIndexSorter(void* extra, const void* l, const void* r)
2306 {
2307 Parser<A>* parser = (Parser<A>*)extra;
2308 const uint32_t* left = (uint32_t*)l;
2309 const uint32_t* right = (uint32_t*)r;
2310 const macho_section<P>* leftSect = parser->machOSectionFromSectionIndex(*left);
2311 const macho_section<P>* rightSect = parser->machOSectionFromSectionIndex(*right);
2312
2313 // can't just return difference because 64-bit diff does not fit in 32-bit return type
2314 int64_t result = leftSect->addr() - rightSect->addr();
2315 if ( result == 0 ) {
2316 // two sections with same start address
2317 // one with zero size goes first
2318 bool leftEmpty = ( leftSect->size() == 0 );
2319 bool rightEmpty = ( rightSect->size() == 0 );
2320 if ( leftEmpty != rightEmpty ) {
2321 return ( rightEmpty ? 1 : -1 );
2322 }
2323 if ( !leftEmpty && !rightEmpty )
2324 throwf("overlapping sections");
2325 // both empty, so chose file order
2326 return ( rightSect - leftSect );
2327 }
2328 else if ( result < 0 )
2329 return -1;
2330 else
2331 return 1;
2332 }
2333
2334 template <typename A>
2335 void Parser<A>::makeSortedSectionsArray(uint32_t array[])
2336 {
2337 const bool log = false;
2338
2339 if ( log ) {
2340 fprintf(stderr, "unsorted sections:\n");
2341 for(unsigned int i=0; i < _machOSectionsCount; ++i )
2342 fprintf(stderr, "0x%08llX %s %s\n", _sectionsStart[i].addr(), _sectionsStart[i].segname(), _sectionsStart[i].sectname());
2343 }
2344
2345 // sort by symbol table address
2346 for (uint32_t i=0; i < _machOSectionsCount; ++i)
2347 array[i] = i;
2348 ::qsort_r(array, _machOSectionsCount, sizeof(uint32_t), this, &sectionIndexSorter);
2349
2350 if ( log ) {
2351 fprintf(stderr, "sorted sections:\n");
2352 for(unsigned int i=0; i < _machOSectionsCount; ++i )
2353 fprintf(stderr, "0x%08llX %s %s\n", _sectionsStart[array[i]].addr(), _sectionsStart[array[i]].segname(), _sectionsStart[array[i]].sectname());
2354 }
2355 }
2356
2357
2358
2359 template <typename A>
2360 int Parser<A>::symbolIndexSorter(void* extra, const void* l, const void* r)
2361 {
2362 ParserAndSectionsArray* extraInfo = (ParserAndSectionsArray*)extra;
2363 Parser<A>* parser = extraInfo->parser;
2364 const uint32_t* sortedSectionsArray = extraInfo->sortedSectionsArray;
2365 const uint32_t* left = (uint32_t*)l;
2366 const uint32_t* right = (uint32_t*)r;
2367 const macho_nlist<P>& leftSym = parser->symbolFromIndex(*left);
2368 const macho_nlist<P>& rightSym = parser->symbolFromIndex(*right);
2369 // can't just return difference because 64-bit diff does not fit in 32-bit return type
2370 int64_t result = leftSym.n_value() - rightSym.n_value();
2371 if ( result == 0 ) {
2372 // two symbols with same address
2373 // if in different sections, sort earlier section first
2374 if ( leftSym.n_sect() != rightSym.n_sect() ) {
2375 for (uint32_t i=0; i < parser->machOSectionCount(); ++i) {
2376 if ( sortedSectionsArray[i]+1 == leftSym.n_sect() )
2377 return -1;
2378 if ( sortedSectionsArray[i]+1 == rightSym.n_sect() )
2379 return 1;
2380 }
2381 }
2382 // two symbols in same section, means one is an alias
2383 // if one is ltmp*, make it an alias (sort first)
2384 const char* leftName = parser->nameFromSymbol(leftSym);
2385 const char* rightName = parser->nameFromSymbol(rightSym);
2386 bool leftIsTmp = strncmp(leftName, "ltmp", 4);
2387 bool rightIsTmp = strncmp(rightName, "ltmp", 4);
2388 if ( leftIsTmp != rightIsTmp ) {
2389 return (rightIsTmp ? -1 : 1);
2390 }
2391
2392 // if only one is global, make the other an alias (sort first)
2393 if ( (leftSym.n_type() & N_EXT) != (rightSym.n_type() & N_EXT) ) {
2394 if ( (rightSym.n_type() & N_EXT) != 0 )
2395 return -1;
2396 else
2397 return 1;
2398 }
2399 // if both are global, sort alphabetically. earlier one will be the alias
2400 return ( strcmp(rightName, leftName) );
2401 }
2402 else if ( result < 0 )
2403 return -1;
2404 else
2405 return 1;
2406 }
2407
2408
2409 template <typename A>
2410 void Parser<A>::makeSortedSymbolsArray(uint32_t array[], const uint32_t sectionArray[])
2411 {
2412 const bool log = false;
2413
2414 uint32_t* p = array;
2415 for (uint32_t i=0; i < this->_symbolCount; ++i) {
2416 const macho_nlist<P>& sym = symbolFromIndex(i);
2417 // ignore stabs
2418 if ( (sym.n_type() & N_STAB) != 0 )
2419 continue;
2420
2421 // only look at definitions
2422 if ( (sym.n_type() & N_TYPE) != N_SECT )
2423 continue;
2424
2425 // 'L' labels do not denote atom breaks
2426 const char* symbolName = this->nameFromSymbol(sym);
2427 if ( symbolName[0] == 'L' )
2428 continue;
2429
2430 // how many def syms in each section
2431 if ( sym.n_sect() > _machOSectionsCount )
2432 throw "bad n_sect in symbol table";
2433
2434 // append to array
2435 *p++ = i;
2436 }
2437 assert(p == &array[_symbolsInSections] && "second pass over symbol table yield a different number of symbols");
2438
2439 // sort by symbol table address
2440 ParserAndSectionsArray extra = { this, sectionArray };
2441 ::qsort_r(array, _symbolsInSections, sizeof(uint32_t), &extra, &symbolIndexSorter);
2442
2443
2444 // look for two symbols at same address
2445 _overlappingSymbols = false;
2446 for (unsigned int i=1; i < _symbolsInSections; ++i) {
2447 if ( symbolFromIndex(array[i-1]).n_value() == symbolFromIndex(array[i]).n_value() ) {
2448 //fprintf(stderr, "overlapping symbols at 0x%08llX\n", symbolFromIndex(array[i-1]).n_value());
2449 _overlappingSymbols = true;
2450 break;
2451 }
2452 }
2453
2454 if ( log ) {
2455 fprintf(stderr, "sorted symbols:\n");
2456 for(unsigned int i=0; i < _symbolsInSections; ++i )
2457 fprintf(stderr, "0x%09llX symIndex=%d sectNum=%2d, %s\n", symbolFromIndex(array[i]).n_value(), array[i], symbolFromIndex(array[i]).n_sect(), nameFromSymbol(symbolFromIndex(array[i])) );
2458 }
2459 }
2460
2461 template <typename A>
2462 void Parser<A>::makeSections()
2463 {
2464 // classify each section by type
2465 // compute how many Section objects will be needed and total size for all
2466 unsigned int totalSectionsSize = 0;
2467 uint8_t machOSectsStorage[sizeof(MachOSectionAndSectionClass<P>)*(_machOSectionsCount+2)]; // also room for tentative-defs and absolute symbols
2468 // allocate raw storage for all section objects on stack
2469 MachOSectionAndSectionClass<P>* machOSects = (MachOSectionAndSectionClass<P>*)machOSectsStorage;
2470 unsigned int count = 0;
2471 // local variable for bitcode parsing
2472 const macho_section<P>* bitcodeSect = NULL;
2473 const macho_section<P>* cmdlineSect = NULL;
2474 const macho_section<P>* swiftCmdlineSect = NULL;
2475 const macho_section<P>* bundleSect = NULL;
2476 bool bitcodeAsm = false;
2477
2478 for (uint32_t i=0; i < _machOSectionsCount; ++i) {
2479 const macho_section<P>* sect = &_sectionsStart[i];
2480 uint8_t sectionType = (sect->flags() & SECTION_TYPE);
2481 if ( (sect->offset() + sect->size() > _fileLength) && (sectionType != S_ZEROFILL) && (sectionType != S_THREAD_LOCAL_ZEROFILL) )
2482 throwf("section %s/%s extends beyond end of file,", sect->segname(), sect->sectname());
2483
2484 if ( (sect->flags() & S_ATTR_DEBUG) != 0 ) {
2485 if ( strcmp(sect->segname(), "__DWARF") == 0 ) {
2486 // note that .o file has dwarf
2487 _file->_debugInfoKind = ld::relocatable::File::kDebugInfoDwarf;
2488 // save off iteresting dwarf sections
2489 if ( strcmp(sect->sectname(), "__debug_info") == 0 )
2490 _file->_dwarfDebugInfoSect = sect;
2491 else if ( strcmp(sect->sectname(), "__debug_abbrev") == 0 )
2492 _file->_dwarfDebugAbbrevSect = sect;
2493 else if ( strcmp(sect->sectname(), "__debug_line") == 0 )
2494 _file->_dwarfDebugLineSect = sect;
2495 else if ( strcmp(sect->sectname(), "__debug_str") == 0 )
2496 _file->_dwarfDebugStringSect = sect;
2497 // linker does not propagate dwarf sections to output file
2498 continue;
2499 }
2500 else if ( strcmp(sect->segname(), "__LD") == 0 ) {
2501 if ( strncmp(sect->sectname(), "__compact_unwind", 16) == 0 ) {
2502 machOSects[count].sect = sect;
2503 totalSectionsSize += sizeof(CUSection<A>);
2504 machOSects[count++].type = sectionTypeCompactUnwind;
2505 continue;
2506 }
2507 }
2508 }
2509 if ( strcmp(sect->segname(), "__LLVM") == 0 ) {
2510 // Process bitcode segement
2511 if ( strncmp(sect->sectname(), "__bitcode", 9) == 0 ) {
2512 bitcodeSect = sect;
2513 } else if ( strncmp(sect->sectname(), "__cmdline", 9) == 0 ) {
2514 cmdlineSect = sect;
2515 } else if ( strncmp(sect->sectname(), "__swift_cmdline", 15) == 0 ) {
2516 swiftCmdlineSect = sect;
2517 } else if ( strncmp(sect->sectname(), "__bundle", 8) == 0 ) {
2518 bundleSect = sect;
2519 } else if ( strncmp(sect->sectname(), "__asm", 5) == 0 ) {
2520 bitcodeAsm = true;
2521 }
2522 // If treat the bitcode as data, continue to parse as a normal section.
2523 if ( !_treateBitcodeAsData )
2524 continue;
2525 }
2526 // ignore empty __OBJC sections
2527 if ( (sect->size() == 0) && (strcmp(sect->segname(), "__OBJC") == 0) )
2528 continue;
2529 // objc image info section is really attributes and not content
2530 if ( ((strcmp(sect->sectname(), "__image_info") == 0) && (strcmp(sect->segname(), "__OBJC") == 0))
2531 || ((strncmp(sect->sectname(), "__objc_imageinfo", 16) == 0) && (strcmp(sect->segname(), "__DATA") == 0)) ) {
2532 // struct objc_image_info {
2533 // uint32_t version; // initially 0
2534 // uint32_t flags;
2535 // };
2536 // #define OBJC_IMAGE_SUPPORTS_GC 2
2537 // #define OBJC_IMAGE_GC_ONLY 4
2538 // #define OBJC_IMAGE_IS_SIMULATED 32
2539 // #define OBJC_IMAGE_HAS_CATEGORY_CLASS_PROPERTIES 64
2540 //
2541 const uint32_t* contents = (uint32_t*)(_file->fileContent()+sect->offset());
2542 if ( (sect->size() >= 8) && (contents[0] == 0) ) {
2543 uint32_t flags = E::get32(contents[1]);
2544 if ( (flags & 4) == 4 )
2545 _file->_objConstraint = ld::File::objcConstraintGC;
2546 else if ( (flags & 2) == 2 )
2547 _file->_objConstraint = ld::File::objcConstraintRetainReleaseOrGC;
2548 else if ( (flags & 32) == 32 )
2549 _file->_objConstraint = ld::File::objcConstraintRetainReleaseForSimulator;
2550 else
2551 _file->_objConstraint = ld::File::objcConstraintRetainRelease;
2552 _file->_swiftVersion = ((flags >> 8) & 0xFF);
2553 _file->_objcHasCategoryClassPropertiesField = (flags & 64);
2554 if ( sect->size() > 8 ) {
2555 warning("section %s/%s has unexpectedly large size %llu in %s",
2556 sect->segname(), Section<A>::makeSectionName(sect), sect->size(), _file->path());
2557 }
2558 }
2559 else {
2560 warning("can't parse %s/%s section in %s", sect->segname(), Section<A>::makeSectionName(sect), _file->path());
2561 }
2562 continue;
2563 }
2564 machOSects[count].sect = sect;
2565 switch ( sect->flags() & SECTION_TYPE ) {
2566 case S_SYMBOL_STUBS:
2567 if ( _stubsSectionNum == 0 ) {
2568 _stubsSectionNum = i+1;
2569 _stubsMachOSection = sect;
2570 }
2571 else
2572 assert(1 && "multiple S_SYMBOL_STUBS sections");
2573 case S_LAZY_SYMBOL_POINTERS:
2574 break;
2575 case S_4BYTE_LITERALS:
2576 totalSectionsSize += sizeof(Literal4Section<A>);
2577 machOSects[count++].type = sectionTypeLiteral4;
2578 break;
2579 case S_8BYTE_LITERALS:
2580 totalSectionsSize += sizeof(Literal8Section<A>);
2581 machOSects[count++].type = sectionTypeLiteral8;
2582 break;
2583 case S_16BYTE_LITERALS:
2584 totalSectionsSize += sizeof(Literal16Section<A>);
2585 machOSects[count++].type = sectionTypeLiteral16;
2586 break;
2587 case S_NON_LAZY_SYMBOL_POINTERS:
2588 totalSectionsSize += sizeof(NonLazyPointerSection<A>);
2589 machOSects[count++].type = sectionTypeNonLazy;
2590 break;
2591 case S_THREAD_LOCAL_VARIABLE_POINTERS:
2592 totalSectionsSize += sizeof(TLVPointerSection<A>);
2593 machOSects[count++].type = sectionTypeTLVPointers;
2594 break;
2595 case S_LITERAL_POINTERS:
2596 if ( (strcmp(sect->segname(), "__OBJC") == 0) && (strcmp(sect->sectname(), "__cls_refs") == 0) ) {
2597 totalSectionsSize += sizeof(Objc1ClassReferences<A>);
2598 machOSects[count++].type = sectionTypeObjC1ClassRefs;
2599 }
2600 else {
2601 totalSectionsSize += sizeof(PointerToCStringSection<A>);
2602 machOSects[count++].type = sectionTypeCStringPointer;
2603 }
2604 break;
2605 case S_CSTRING_LITERALS:
2606 totalSectionsSize += sizeof(CStringSection<A>);
2607 machOSects[count++].type = sectionTypeCString;
2608 break;
2609 case S_MOD_INIT_FUNC_POINTERS:
2610 case S_MOD_TERM_FUNC_POINTERS:
2611 case S_THREAD_LOCAL_INIT_FUNCTION_POINTERS:
2612 case S_INTERPOSING:
2613 case S_ZEROFILL:
2614 case S_REGULAR:
2615 case S_COALESCED:
2616 case S_THREAD_LOCAL_REGULAR:
2617 case S_THREAD_LOCAL_ZEROFILL:
2618 if ( (strcmp(sect->segname(), "__TEXT") == 0) && (strcmp(sect->sectname(), "__eh_frame") == 0) ) {
2619 totalSectionsSize += sizeof(CFISection<A>);
2620 machOSects[count++].type = sectionTypeCFI;
2621 }
2622 else if ( (strcmp(sect->segname(), "__DATA") == 0) && (strcmp(sect->sectname(), "__cfstring") == 0) ) {
2623 totalSectionsSize += sizeof(CFStringSection<A>);
2624 machOSects[count++].type = sectionTypeCFString;
2625 }
2626 else if ( (strcmp(sect->segname(), "__TEXT") == 0) && (strcmp(sect->sectname(), "__ustring") == 0) ) {
2627 totalSectionsSize += sizeof(UTF16StringSection<A>);
2628 machOSects[count++].type = sectionTypeUTF16Strings;
2629 }
2630 else if ( (strcmp(sect->segname(), "__DATA") == 0) && (strncmp(sect->sectname(), "__objc_classrefs", 16) == 0) ) {
2631 totalSectionsSize += sizeof(ObjC2ClassRefsSection<A>);
2632 machOSects[count++].type = sectionTypeObjC2ClassRefs;
2633 }
2634 else if ( (strcmp(sect->segname(), "__DATA") == 0) && (strcmp(sect->sectname(), "__objc_catlist") == 0) ) {
2635 totalSectionsSize += sizeof(ObjC2CategoryListSection<A>);
2636 machOSects[count++].type = typeObjC2CategoryList;
2637 }
2638 else if ( _AppleObjc && (strcmp(sect->segname(), "__OBJC") == 0) && (strcmp(sect->sectname(), "__class") == 0) ) {
2639 totalSectionsSize += sizeof(ObjC1ClassSection<A>);
2640 machOSects[count++].type = sectionTypeObjC1Classes;
2641 }
2642 else {
2643 totalSectionsSize += sizeof(SymboledSection<A>);
2644 machOSects[count++].type = sectionTypeSymboled;
2645 }
2646 break;
2647 case S_THREAD_LOCAL_VARIABLES:
2648 totalSectionsSize += sizeof(TLVDefsSection<A>);
2649 machOSects[count++].type = sectionTypeTLVDefs;
2650 break;
2651 default:
2652 throwf("unknown section type %d", sect->flags() & SECTION_TYPE);
2653 }
2654 }
2655
2656 // Create bitcode
2657 if ( bitcodeSect != NULL ) {
2658 if ( cmdlineSect != NULL )
2659 _file->_bitcode = std::unique_ptr<ld::Bitcode>(new ld::ClangBitcode(&_fileContent[bitcodeSect->offset()], bitcodeSect->size(),
2660 &_fileContent[cmdlineSect->offset()], cmdlineSect->size()));
2661 else if ( swiftCmdlineSect != NULL )
2662 _file->_bitcode = std::unique_ptr<ld::Bitcode>(new ld::SwiftBitcode(&_fileContent[bitcodeSect->offset()], bitcodeSect->size(),
2663 &_fileContent[swiftCmdlineSect->offset()], swiftCmdlineSect->size()));
2664 else
2665 throwf("Object file with bitcode missing cmdline options: %s", _file->path());
2666 }
2667 else if ( bundleSect != NULL )
2668 _file->_bitcode = std::unique_ptr<ld::Bitcode>(new ld::BundleBitcode(&_fileContent[bundleSect->offset()], bundleSect->size()));
2669 else if ( bitcodeAsm )
2670 _file->_bitcode = std::unique_ptr<ld::Bitcode>(new ld::AsmBitcode(_fileContent, _fileLength));
2671
2672 // sort by address (mach-o object files don't aways have sections sorted)
2673 ::qsort(machOSects, count, sizeof(MachOSectionAndSectionClass<P>), MachOSectionAndSectionClass<P>::sorter);
2674
2675 // we will synthesize a dummy Section<A> object for tentative definitions
2676 if ( _tentativeDefinitionCount > 0 ) {
2677 totalSectionsSize += sizeof(TentativeDefinitionSection<A>);
2678 machOSects[count++].type = sectionTypeTentativeDefinitions;
2679 }
2680
2681 // we will synthesize a dummy Section<A> object for Absolute symbols
2682 if ( _absoluteSymbolCount > 0 ) {
2683 totalSectionsSize += sizeof(AbsoluteSymbolSection<A>);
2684 machOSects[count++].type = sectionTypeAbsoluteSymbols;
2685 }
2686
2687 // allocate one block for all Section objects as well as pointers to each
2688 uint8_t* space = new uint8_t[totalSectionsSize+count*sizeof(Section<A>*)];
2689 _file->_sectionsArray = (Section<A>**)space;
2690 _file->_sectionsArrayCount = count;
2691 Section<A>** objects = _file->_sectionsArray;
2692 space += count*sizeof(Section<A>*);
2693 for (uint32_t i=0; i < count; ++i) {
2694 switch ( machOSects[i].type ) {
2695 case sectionTypeIgnore:
2696 break;
2697 case sectionTypeLiteral4:
2698 *objects++ = new (space) Literal4Section<A>(*this, *_file, machOSects[i].sect);
2699 space += sizeof(Literal4Section<A>);
2700 break;
2701 case sectionTypeLiteral8:
2702 *objects++ = new (space) Literal8Section<A>(*this, *_file, machOSects[i].sect);
2703 space += sizeof(Literal8Section<A>);
2704 break;
2705 case sectionTypeLiteral16:
2706 *objects++ = new (space) Literal16Section<A>(*this, *_file, machOSects[i].sect);
2707 space += sizeof(Literal16Section<A>);
2708 break;
2709 case sectionTypeNonLazy:
2710 *objects++ = new (space) NonLazyPointerSection<A>(*this, *_file, machOSects[i].sect);
2711 space += sizeof(NonLazyPointerSection<A>);
2712 break;
2713 case sectionTypeTLVPointers:
2714 *objects++ = new (space) TLVPointerSection<A>(*this, *_file, machOSects[i].sect);
2715 space += sizeof(TLVPointerSection<A>);
2716 break;
2717 case sectionTypeCFI:
2718 _EHFrameSection = new (space) CFISection<A>(*this, *_file, machOSects[i].sect);
2719 *objects++ = _EHFrameSection;
2720 space += sizeof(CFISection<A>);
2721 break;
2722 case sectionTypeCString:
2723 *objects++ = new (space) CStringSection<A>(*this, *_file, machOSects[i].sect);
2724 space += sizeof(CStringSection<A>);
2725 break;
2726 case sectionTypeCStringPointer:
2727 *objects++ = new (space) PointerToCStringSection<A>(*this, *_file, machOSects[i].sect);
2728 space += sizeof(PointerToCStringSection<A>);
2729 break;
2730 case sectionTypeObjC1ClassRefs:
2731 *objects++ = new (space) Objc1ClassReferences<A>(*this, *_file, machOSects[i].sect);
2732 space += sizeof(Objc1ClassReferences<A>);
2733 break;
2734 case sectionTypeUTF16Strings:
2735 *objects++ = new (space) UTF16StringSection<A>(*this, *_file, machOSects[i].sect);
2736 space += sizeof(UTF16StringSection<A>);
2737 break;
2738 case sectionTypeCFString:
2739 *objects++ = new (space) CFStringSection<A>(*this, *_file, machOSects[i].sect);
2740 space += sizeof(CFStringSection<A>);
2741 break;
2742 case sectionTypeObjC2ClassRefs:
2743 *objects++ = new (space) ObjC2ClassRefsSection<A>(*this, *_file, machOSects[i].sect);
2744 space += sizeof(ObjC2ClassRefsSection<A>);
2745 break;
2746 case typeObjC2CategoryList:
2747 *objects++ = new (space) ObjC2CategoryListSection<A>(*this, *_file, machOSects[i].sect);
2748 space += sizeof(ObjC2CategoryListSection<A>);
2749 break;
2750 case sectionTypeObjC1Classes:
2751 *objects++ = new (space) ObjC1ClassSection<A>(*this, *_file, machOSects[i].sect);
2752 space += sizeof(ObjC1ClassSection<A>);
2753 break;
2754 case sectionTypeSymboled:
2755 *objects++ = new (space) SymboledSection<A>(*this, *_file, machOSects[i].sect);
2756 space += sizeof(SymboledSection<A>);
2757 break;
2758 case sectionTypeTLVDefs:
2759 *objects++ = new (space) TLVDefsSection<A>(*this, *_file, machOSects[i].sect);
2760 space += sizeof(TLVDefsSection<A>);
2761 break;
2762 case sectionTypeCompactUnwind:
2763 _compactUnwindSection = new (space) CUSection<A>(*this, *_file, machOSects[i].sect);
2764 *objects++ = _compactUnwindSection;
2765 space += sizeof(CUSection<A>);
2766 break;
2767 case sectionTypeTentativeDefinitions:
2768 *objects++ = new (space) TentativeDefinitionSection<A>(*this, *_file);
2769 space += sizeof(TentativeDefinitionSection<A>);
2770 break;
2771 case sectionTypeAbsoluteSymbols:
2772 _absoluteSection = new (space) AbsoluteSymbolSection<A>(*this, *_file);
2773 *objects++ = _absoluteSection;
2774 space += sizeof(AbsoluteSymbolSection<A>);
2775 break;
2776 default:
2777 throw "internal error uknown SectionType";
2778 }
2779 }
2780 }
2781
2782
2783 template <typename A>
2784 Section<A>* Parser<A>::sectionForAddress(typename A::P::uint_t addr)
2785 {
2786 for (uint32_t i=0; i < _file->_sectionsArrayCount; ++i ) {
2787 const macho_section<typename A::P>* sect = _file->_sectionsArray[i]->machoSection();
2788 // TentativeDefinitionSection and AbsoluteSymbolSection have no mach-o section
2789 if ( sect != NULL ) {
2790 if ( (sect->addr() <= addr) && (addr < (sect->addr()+sect->size())) ) {
2791 return _file->_sectionsArray[i];
2792 }
2793 }
2794 }
2795 // not strictly in any section
2796 // may be in a zero length section
2797 for (uint32_t i=0; i < _file->_sectionsArrayCount; ++i ) {
2798 const macho_section<typename A::P>* sect = _file->_sectionsArray[i]->machoSection();
2799 // TentativeDefinitionSection and AbsoluteSymbolSection have no mach-o section
2800 if ( sect != NULL ) {
2801 if ( (sect->addr() == addr) && (sect->size() == 0) ) {
2802 return _file->_sectionsArray[i];
2803 }
2804 }
2805 }
2806
2807 throwf("sectionForAddress(0x%llX) address not in any section", (uint64_t)addr);
2808 }
2809
2810 template <typename A>
2811 Section<A>* Parser<A>::sectionForNum(unsigned int num)
2812 {
2813 for (uint32_t i=0; i < _file->_sectionsArrayCount; ++i ) {
2814 const macho_section<typename A::P>* sect = _file->_sectionsArray[i]->machoSection();
2815 // TentativeDefinitionSection and AbsoluteSymbolSection have no mach-o section
2816 if ( sect != NULL ) {
2817 if ( num == (unsigned int)((sect - _sectionsStart)+1) )
2818 return _file->_sectionsArray[i];
2819 }
2820 }
2821 throwf("sectionForNum(%u) section number not for any section", num);
2822 }
2823
2824 template <typename A>
2825 Atom<A>* Parser<A>::findAtomByAddress(pint_t addr)
2826 {
2827 Section<A>* section = this->sectionForAddress(addr);
2828 return section->findAtomByAddress(addr);
2829 }
2830
2831 template <typename A>
2832 Atom<A>* Parser<A>::findAtomByAddressOrNullIfStub(pint_t addr)
2833 {
2834 if ( hasStubsSection() && (_stubsMachOSection->addr() <= addr) && (addr < (_stubsMachOSection->addr()+_stubsMachOSection->size())) )
2835 return NULL;
2836 return findAtomByAddress(addr);
2837 }
2838
2839 template <typename A>
2840 Atom<A>* Parser<A>::findAtomByAddressOrLocalTargetOfStub(pint_t addr, uint32_t* offsetInAtom)
2841 {
2842 if ( hasStubsSection() && (_stubsMachOSection->addr() <= addr) && (addr < (_stubsMachOSection->addr()+_stubsMachOSection->size())) ) {
2843 // target is a stub, remove indirection
2844 uint32_t symbolIndex = this->symbolIndexFromIndirectSectionAddress(addr, _stubsMachOSection);
2845 assert(symbolIndex != INDIRECT_SYMBOL_LOCAL);
2846 const macho_nlist<P>& sym = this->symbolFromIndex(symbolIndex);
2847 // can't be to external weak symbol
2848 assert( (this->combineFromSymbol(sym) != ld::Atom::combineByName) || (this->scopeFromSymbol(sym) != ld::Atom::scopeGlobal) );
2849 *offsetInAtom = 0;
2850 return this->findAtomByName(this->nameFromSymbol(sym));
2851 }
2852 Atom<A>* target = this->findAtomByAddress(addr);
2853 *offsetInAtom = addr - target->_objAddress;
2854 return target;
2855 }
2856
2857 template <typename A>
2858 Atom<A>* Parser<A>::findAtomByName(const char* name)
2859 {
2860 uint8_t* p = _file->_atomsArray;
2861 for(int i=_file->_atomsArrayCount; i > 0; --i) {
2862 Atom<A>* atom = (Atom<A>*)p;
2863 if ( strcmp(name, atom->name()) == 0 )
2864 return atom;
2865 p += sizeof(Atom<A>);
2866 }
2867 return NULL;
2868 }
2869
2870 template <typename A>
2871 void Parser<A>::findTargetFromAddress(pint_t addr, TargetDesc& target)
2872 {
2873 if ( hasStubsSection() && (_stubsMachOSection->addr() <= addr) && (addr < (_stubsMachOSection->addr()+_stubsMachOSection->size())) ) {
2874 // target is a stub, remove indirection
2875 uint32_t symbolIndex = this->symbolIndexFromIndirectSectionAddress(addr, _stubsMachOSection);
2876 assert(symbolIndex != INDIRECT_SYMBOL_LOCAL);
2877 const macho_nlist<P>& sym = this->symbolFromIndex(symbolIndex);
2878 target.atom = NULL;
2879 target.name = this->nameFromSymbol(sym);
2880 target.weakImport = this->weakImportFromSymbol(sym);
2881 target.addend = 0;
2882 return;
2883 }
2884 Section<A>* section = this->sectionForAddress(addr);
2885 target.atom = section->findAtomByAddress(addr);
2886 target.addend = addr - target.atom->_objAddress;
2887 target.weakImport = false;
2888 target.name = NULL;
2889 }
2890
2891 template <typename A>
2892 void Parser<A>::findTargetFromAddress(pint_t baseAddr, pint_t addr, TargetDesc& target)
2893 {
2894 findTargetFromAddress(baseAddr, target);
2895 target.addend = addr - target.atom->_objAddress;
2896 }
2897
2898 template <typename A>
2899 void Parser<A>::findTargetFromAddressAndSectionNum(pint_t addr, unsigned int sectNum, TargetDesc& target)
2900 {
2901 if ( sectNum == R_ABS ) {
2902 // target is absolute symbol that corresponds to addr
2903 if ( _absoluteSection != NULL ) {
2904 target.atom = _absoluteSection->findAbsAtomForValue(addr);
2905 if ( target.atom != NULL ) {
2906 target.name = NULL;
2907 target.weakImport = false;
2908 target.addend = 0;
2909 return;
2910 }
2911 }
2912 throwf("R_ABS reloc but no absolute symbol at target address");
2913 }
2914
2915 if ( hasStubsSection() && (stubsSectionNum() == sectNum) ) {
2916 // target is a stub, remove indirection
2917 uint32_t symbolIndex = this->symbolIndexFromIndirectSectionAddress(addr, _stubsMachOSection);
2918 assert(symbolIndex != INDIRECT_SYMBOL_LOCAL);
2919 const macho_nlist<P>& sym = this->symbolFromIndex(symbolIndex);
2920 // use direct reference when stub is to a static function
2921 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (this->nameFromSymbol(sym)[0] == 'L')) ) {
2922 this->findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
2923 }
2924 else {
2925 target.atom = NULL;
2926 target.name = this->nameFromSymbol(sym);
2927 target.weakImport = this->weakImportFromSymbol(sym);
2928 target.addend = 0;
2929 }
2930 return;
2931 }
2932 Section<A>* section = this->sectionForNum(sectNum);
2933 target.atom = section->findAtomByAddress(addr);
2934 if ( target.atom == NULL ) {
2935 typedef typename A::P::sint_t sint_t;
2936 sint_t a = (sint_t)addr;
2937 sint_t sectStart = (sint_t)(section->machoSection()->addr());
2938 sint_t sectEnd = sectStart + section->machoSection()->size();
2939 if ( a < sectStart ) {
2940 // target address is before start of section, so must be negative addend
2941 target.atom = section->findAtomByAddress(sectStart);
2942 target.addend = a - sectStart;
2943 target.weakImport = false;
2944 target.name = NULL;
2945 return;
2946 }
2947 else if ( a >= sectEnd ) {
2948 target.atom = section->findAtomByAddress(sectEnd-1);
2949 target.addend = a - sectEnd;
2950 target.weakImport = false;
2951 target.name = NULL;
2952 return;
2953 }
2954 }
2955 assert(target.atom != NULL);
2956 target.addend = addr - target.atom->_objAddress;
2957 target.weakImport = false;
2958 target.name = NULL;
2959 }
2960
2961 template <typename A>
2962 void Parser<A>::addDtraceExtraInfos(const SourceLocation& src, const char* providerName)
2963 {
2964 // for every ___dtrace_stability$* and ___dtrace_typedefs$* undefine with
2965 // a matching provider name, add a by-name kDtraceTypeReference at probe site
2966 const char* dollar = strchr(providerName, '$');
2967 if ( dollar != NULL ) {
2968 int providerNameLen = dollar-providerName+1;
2969 for ( std::vector<const char*>::iterator it = _dtraceProviderInfo.begin(); it != _dtraceProviderInfo.end(); ++it) {
2970 const char* typeDollar = strchr(*it, '$');
2971 if ( typeDollar != NULL ) {
2972 if ( strncmp(typeDollar+1, providerName, providerNameLen) == 0 ) {
2973 addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindDtraceExtra,false, *it);
2974 }
2975 }
2976 }
2977 }
2978 }
2979
2980 template <typename A>
2981 const char* Parser<A>::scanSymbolTableForAddress(uint64_t addr)
2982 {
2983 uint64_t closestSymAddr = 0;
2984 const char* closestSymName = NULL;
2985 for (uint32_t i=0; i < this->_symbolCount; ++i) {
2986 const macho_nlist<P>& sym = symbolFromIndex(i);
2987 // ignore stabs
2988 if ( (sym.n_type() & N_STAB) != 0 )
2989 continue;
2990
2991 // only look at definitions
2992 if ( (sym.n_type() & N_TYPE) != N_SECT )
2993 continue;
2994
2995 // return with exact match
2996 if ( sym.n_value() == addr ) {
2997 const char* name = nameFromSymbol(sym);
2998 if ( strncmp(name, "ltmp", 4) != 0 )
2999 return name;
3000 // treat 'ltmp*' labels as close match
3001 closestSymAddr = sym.n_value();
3002 closestSymName = name;
3003 }
3004
3005 // record closest seen so far
3006 if ( (sym.n_value() < addr) && ((sym.n_value() > closestSymAddr) || (closestSymName == NULL)) )
3007 closestSymName = nameFromSymbol(sym);
3008 }
3009
3010 return (closestSymName != NULL) ? closestSymName : "unknown";
3011 }
3012
3013
3014 template <typename A>
3015 void Parser<A>::addFixups(const SourceLocation& src, ld::Fixup::Kind setKind, const TargetDesc& target)
3016 {
3017 // some fixup pairs can be combined
3018 ld::Fixup::Cluster cl = ld::Fixup::k1of3;
3019 ld::Fixup::Kind firstKind = ld::Fixup::kindSetTargetAddress;
3020 bool combined = false;
3021 if ( target.addend == 0 ) {
3022 cl = ld::Fixup::k1of1;
3023 combined = true;
3024 switch ( setKind ) {
3025 case ld::Fixup::kindStoreLittleEndian32:
3026 firstKind = ld::Fixup::kindStoreTargetAddressLittleEndian32;
3027 break;
3028 case ld::Fixup::kindStoreLittleEndian64:
3029 firstKind = ld::Fixup::kindStoreTargetAddressLittleEndian64;
3030 break;
3031 case ld::Fixup::kindStoreBigEndian32:
3032 firstKind = ld::Fixup::kindStoreTargetAddressBigEndian32;
3033 break;
3034 case ld::Fixup::kindStoreBigEndian64:
3035 firstKind = ld::Fixup::kindStoreTargetAddressBigEndian64;
3036 break;
3037 case ld::Fixup::kindStoreX86BranchPCRel32:
3038 firstKind = ld::Fixup::kindStoreTargetAddressX86BranchPCRel32;
3039 break;
3040 case ld::Fixup::kindStoreX86PCRel32:
3041 firstKind = ld::Fixup::kindStoreTargetAddressX86PCRel32;
3042 break;
3043 case ld::Fixup::kindStoreX86PCRel32GOTLoad:
3044 firstKind = ld::Fixup::kindStoreTargetAddressX86PCRel32GOTLoad;
3045 break;
3046 case ld::Fixup::kindStoreX86PCRel32TLVLoad:
3047 firstKind = ld::Fixup::kindStoreTargetAddressX86PCRel32TLVLoad;
3048 break;
3049 case ld::Fixup::kindStoreX86Abs32TLVLoad:
3050 firstKind = ld::Fixup::kindStoreTargetAddressX86Abs32TLVLoad;
3051 break;
3052 case ld::Fixup::kindStoreARMBranch24:
3053 firstKind = ld::Fixup::kindStoreTargetAddressARMBranch24;
3054 break;
3055 case ld::Fixup::kindStoreThumbBranch22:
3056 firstKind = ld::Fixup::kindStoreTargetAddressThumbBranch22;
3057 break;
3058 #if SUPPORT_ARCH_arm64
3059 case ld::Fixup::kindStoreARM64Branch26:
3060 firstKind = ld::Fixup::kindStoreTargetAddressARM64Branch26;
3061 break;
3062 case ld::Fixup::kindStoreARM64Page21:
3063 firstKind = ld::Fixup::kindStoreTargetAddressARM64Page21;
3064 break;
3065 case ld::Fixup::kindStoreARM64PageOff12:
3066 firstKind = ld::Fixup::kindStoreTargetAddressARM64PageOff12;
3067 break;
3068 case ld::Fixup::kindStoreARM64GOTLoadPage21:
3069 firstKind = ld::Fixup::kindStoreTargetAddressARM64GOTLoadPage21;
3070 break;
3071 case ld::Fixup::kindStoreARM64GOTLoadPageOff12:
3072 firstKind = ld::Fixup::kindStoreTargetAddressARM64GOTLoadPageOff12;
3073 break;
3074 case ld::Fixup::kindStoreARM64TLVPLoadPage21:
3075 firstKind = ld::Fixup::kindStoreTargetAddressARM64TLVPLoadPage21;
3076 break;
3077 case ld::Fixup::kindStoreARM64TLVPLoadPageOff12:
3078 firstKind = ld::Fixup::kindStoreTargetAddressARM64TLVPLoadPageOff12;
3079 break;
3080 #endif
3081 default:
3082 combined = false;
3083 cl = ld::Fixup::k1of2;
3084 break;
3085 }
3086 }
3087
3088 if ( target.atom != NULL ) {
3089 if ( target.atom->scope() == ld::Atom::scopeTranslationUnit ) {
3090 addFixup(src, cl, firstKind, target.atom);
3091 }
3092 else if ( (target.atom->combine() == ld::Atom::combineByNameAndContent) || (target.atom->combine() == ld::Atom::combineByNameAndReferences) ) {
3093 addFixup(src, cl, firstKind, ld::Fixup::bindingByContentBound, target.atom);
3094 }
3095 else if ( (src.atom->section().type() == ld::Section::typeCFString) && (src.offsetInAtom != 0) ) {
3096 // backing string in CFStrings should always be direct
3097 addFixup(src, cl, firstKind, target.atom);
3098 }
3099 else if ( (src.atom == target.atom) && (target.atom->combine() == ld::Atom::combineByName) ) {
3100 // reference to self should always be direct
3101 addFixup(src, cl, firstKind, target.atom);
3102 }
3103 else {
3104 // change direct fixup to by-name fixup
3105 addFixup(src, cl, firstKind, false, target.atom->name());
3106 }
3107 }
3108 else {
3109 addFixup(src, cl, firstKind, target.weakImport, target.name);
3110 }
3111 if ( target.addend == 0 ) {
3112 if ( ! combined )
3113 addFixup(src, ld::Fixup::k2of2, setKind);
3114 }
3115 else {
3116 addFixup(src, ld::Fixup::k2of3, ld::Fixup::kindAddAddend, target.addend);
3117 addFixup(src, ld::Fixup::k3of3, setKind);
3118 }
3119 }
3120
3121 template <typename A>
3122 void Parser<A>::addFixups(const SourceLocation& src, ld::Fixup::Kind kind, const TargetDesc& target, const TargetDesc& picBase)
3123 {
3124 ld::Fixup::Cluster cl = (target.addend == 0) ? ld::Fixup::k1of4 : ld::Fixup::k1of5;
3125 if ( target.atom != NULL ) {
3126 if ( target.atom->scope() == ld::Atom::scopeTranslationUnit ) {
3127 addFixup(src, cl, ld::Fixup::kindSetTargetAddress, target.atom);
3128 }
3129 else if ( (target.atom->combine() == ld::Atom::combineByNameAndContent) || (target.atom->combine() == ld::Atom::combineByNameAndReferences) ) {
3130 addFixup(src, cl, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, target.atom);
3131 }
3132 else {
3133 addFixup(src, cl, ld::Fixup::kindSetTargetAddress, false, target.atom->name());
3134 }
3135 }
3136 else {
3137 addFixup(src, cl, ld::Fixup::kindSetTargetAddress, target.weakImport, target.name);
3138 }
3139 if ( target.addend == 0 ) {
3140 assert(picBase.atom != NULL);
3141 addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, picBase.atom);
3142 addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, picBase.addend);
3143 addFixup(src, ld::Fixup::k4of4, kind);
3144 }
3145 else {
3146 addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, target.addend);
3147 addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, picBase.atom);
3148 addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, picBase.addend);
3149 addFixup(src, ld::Fixup::k5of5, kind);
3150 }
3151 }
3152
3153
3154
3155 template <typename A>
3156 uint32_t TentativeDefinitionSection<A>::computeAtomCount(class Parser<A>& parser,
3157 struct Parser<A>::LabelAndCFIBreakIterator& it,
3158 const struct Parser<A>::CFI_CU_InfoArrays&)
3159 {
3160 return parser.tentativeDefinitionCount();
3161 }
3162
3163 template <typename A>
3164 uint32_t TentativeDefinitionSection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
3165 struct Parser<A>::LabelAndCFIBreakIterator& it,
3166 const struct Parser<A>::CFI_CU_InfoArrays&)
3167 {
3168 this->_beginAtoms = (Atom<A>*)p;
3169 uint32_t count = 0;
3170 for (uint32_t i=parser.undefinedStartIndex(); i < parser.undefinedEndIndex(); ++i) {
3171 const macho_nlist<P>& sym = parser.symbolFromIndex(i);
3172 if ( ((sym.n_type() & N_TYPE) == N_UNDF) && (sym.n_value() != 0) ) {
3173 uint64_t size = sym.n_value();
3174 uint8_t alignP2 = GET_COMM_ALIGN(sym.n_desc());
3175 if ( alignP2 == 0 ) {
3176 // common symbols align to their size
3177 // that is, a 4-byte common aligns to 4-bytes
3178 // if this size is not a power of two,
3179 // then round up to the next power of two
3180 alignP2 = 63 - (uint8_t)__builtin_clzll(size);
3181 if ( size != (1ULL << alignP2) )
3182 ++alignP2;
3183 // <rdar://problem/24871389> limit default alignment of large commons
3184 if ( alignP2 > parser.maxDefaultCommonAlignment() )
3185 alignP2 = parser.maxDefaultCommonAlignment();
3186 }
3187 Atom<A>* allocatedSpace = (Atom<A>*)p;
3188 new (allocatedSpace) Atom<A>(*this, parser.nameFromSymbol(sym), (pint_t)ULLONG_MAX, size,
3189 ld::Atom::definitionTentative, ld::Atom::combineByName,
3190 parser.scopeFromSymbol(sym), ld::Atom::typeZeroFill, ld::Atom::symbolTableIn,
3191 parser.dontDeadStripFromSymbol(sym), false, false, ld::Atom::Alignment(alignP2) );
3192 p += sizeof(Atom<A>);
3193 ++count;
3194 }
3195 }
3196 this->_endAtoms = (Atom<A>*)p;
3197 return count;
3198 }
3199
3200
3201 template <typename A>
3202 uint32_t AbsoluteSymbolSection<A>::computeAtomCount(class Parser<A>& parser,
3203 struct Parser<A>::LabelAndCFIBreakIterator& it,
3204 const struct Parser<A>::CFI_CU_InfoArrays&)
3205 {
3206 return parser.absoluteSymbolCount();
3207 }
3208
3209 template <typename A>
3210 uint32_t AbsoluteSymbolSection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
3211 struct Parser<A>::LabelAndCFIBreakIterator& it,
3212 const struct Parser<A>::CFI_CU_InfoArrays&)
3213 {
3214 this->_beginAtoms = (Atom<A>*)p;
3215 uint32_t count = 0;
3216 for (uint32_t i=0; i < parser.symbolCount(); ++i) {
3217 const macho_nlist<P>& sym = parser.symbolFromIndex(i);
3218 if ( (sym.n_type() & N_TYPE) != N_ABS )
3219 continue;
3220 const char* absName = parser.nameFromSymbol(sym);
3221 // ignore .objc_class_name_* symbols
3222 if ( strncmp(absName, ".objc_class_name_", 17) == 0 )
3223 continue;
3224 // ignore .objc_class_name_* symbols
3225 if ( strncmp(absName, ".objc_category_name_", 20) == 0 )
3226 continue;
3227 // ignore empty *.eh symbols
3228 if ( strcmp(&absName[strlen(absName)-3], ".eh") == 0 )
3229 continue;
3230
3231 Atom<A>* allocatedSpace = (Atom<A>*)p;
3232 new (allocatedSpace) Atom<A>(*this, parser, sym, 0);
3233 p += sizeof(Atom<A>);
3234 ++count;
3235 }
3236 this->_endAtoms = (Atom<A>*)p;
3237 return count;
3238 }
3239
3240 template <typename A>
3241 Atom<A>* AbsoluteSymbolSection<A>::findAbsAtomForValue(typename A::P::uint_t value)
3242 {
3243 Atom<A>* end = this->_endAtoms;
3244 for(Atom<A>* p = this->_beginAtoms; p < end; ++p) {
3245 if ( p->_objAddress == value )
3246 return p;
3247 }
3248 return NULL;
3249 }
3250
3251
3252 template <typename A>
3253 uint32_t Parser<A>::indirectSymbol(uint32_t indirectIndex)
3254 {
3255 if ( indirectIndex >= _indirectTableCount )
3256 throw "indirect symbol index out of range";
3257 return E::get32(_indirectTable[indirectIndex]);
3258 }
3259
3260 template <typename A>
3261 const macho_nlist<typename A::P>& Parser<A>::symbolFromIndex(uint32_t index)
3262 {
3263 if ( index > _symbolCount )
3264 throw "symbol index out of range";
3265 return _symbols[index];
3266 }
3267
3268 template <typename A>
3269 const macho_section<typename A::P>* Parser<A>::machOSectionFromSectionIndex(uint32_t index)
3270 {
3271 if ( index >= _machOSectionsCount )
3272 throw "section index out of range";
3273 return &_sectionsStart[index];
3274 }
3275
3276 template <typename A>
3277 uint32_t Parser<A>::symbolIndexFromIndirectSectionAddress(pint_t addr, const macho_section<P>* sect)
3278 {
3279 uint32_t elementSize = 0;
3280 switch ( sect->flags() & SECTION_TYPE ) {
3281 case S_SYMBOL_STUBS:
3282 elementSize = sect->reserved2();
3283 break;
3284 case S_LAZY_SYMBOL_POINTERS:
3285 case S_NON_LAZY_SYMBOL_POINTERS:
3286 case S_THREAD_LOCAL_VARIABLE_POINTERS:
3287 elementSize = sizeof(pint_t);
3288 break;
3289 default:
3290 throw "section does not use indirect symbol table";
3291 }
3292 uint32_t indexInSection = (addr - sect->addr()) / elementSize;
3293 uint32_t indexIntoIndirectTable = sect->reserved1() + indexInSection;
3294 return this->indirectSymbol(indexIntoIndirectTable);
3295 }
3296
3297
3298
3299 template <typename A>
3300 const char* Parser<A>::nameFromSymbol(const macho_nlist<P>& sym)
3301 {
3302 return &_strings[sym.n_strx()];
3303 }
3304
3305 template <typename A>
3306 ld::Atom::Scope Parser<A>::scopeFromSymbol(const macho_nlist<P>& sym)
3307 {
3308 if ( (sym.n_type() & N_EXT) == 0 )
3309 return ld::Atom::scopeTranslationUnit;
3310 else if ( (sym.n_type() & N_PEXT) != 0 )
3311 return ld::Atom::scopeLinkageUnit;
3312 else if ( this->nameFromSymbol(sym)[0] == 'l' ) // since all 'l' symbols will be remove, don't make them global
3313 return ld::Atom::scopeLinkageUnit;
3314 else
3315 return ld::Atom::scopeGlobal;
3316 }
3317
3318 template <typename A>
3319 ld::Atom::Definition Parser<A>::definitionFromSymbol(const macho_nlist<P>& sym)
3320 {
3321 switch ( sym.n_type() & N_TYPE ) {
3322 case N_ABS:
3323 return ld::Atom::definitionAbsolute;
3324 case N_SECT:
3325 return ld::Atom::definitionRegular;
3326 case N_UNDF:
3327 if ( sym.n_value() != 0 )
3328 return ld::Atom::definitionTentative;
3329 }
3330 throw "definitionFromSymbol() bad symbol";
3331 }
3332
3333 template <typename A>
3334 ld::Atom::Combine Parser<A>::combineFromSymbol(const macho_nlist<P>& sym)
3335 {
3336 if ( sym.n_desc() & N_WEAK_DEF )
3337 return ld::Atom::combineByName;
3338 else
3339 return ld::Atom::combineNever;
3340 }
3341
3342
3343 template <typename A>
3344 ld::Atom::SymbolTableInclusion Parser<A>::inclusionFromSymbol(const macho_nlist<P>& sym)
3345 {
3346 const char* symbolName = nameFromSymbol(sym);
3347 // labels beginning with 'l' (lowercase ell) are automatically removed in final linked images <rdar://problem/4571042>
3348 // labels beginning with 'L' should have been stripped by the assembler, so are stripped now
3349 if ( sym.n_desc() & REFERENCED_DYNAMICALLY )
3350 return ld::Atom::symbolTableInAndNeverStrip;
3351 else if ( symbolName[0] == 'l' )
3352 return ld::Atom::symbolTableNotInFinalLinkedImages;
3353 else if ( symbolName[0] == 'L' )
3354 return ld::Atom::symbolTableNotIn;
3355 else
3356 return ld::Atom::symbolTableIn;
3357 }
3358
3359 template <typename A>
3360 bool Parser<A>::dontDeadStripFromSymbol(const macho_nlist<P>& sym)
3361 {
3362 return ( (sym.n_desc() & (N_NO_DEAD_STRIP|REFERENCED_DYNAMICALLY)) != 0 );
3363 }
3364
3365 template <typename A>
3366 bool Parser<A>::isThumbFromSymbol(const macho_nlist<P>& sym)
3367 {
3368 return ( sym.n_desc() & N_ARM_THUMB_DEF );
3369 }
3370
3371 template <typename A>
3372 bool Parser<A>::weakImportFromSymbol(const macho_nlist<P>& sym)
3373 {
3374 return ( ((sym.n_type() & N_TYPE) == N_UNDF) && ((sym.n_desc() & N_WEAK_REF) != 0) );
3375 }
3376
3377 template <typename A>
3378 bool Parser<A>::resolverFromSymbol(const macho_nlist<P>& sym)
3379 {
3380 return ( sym.n_desc() & N_SYMBOL_RESOLVER );
3381 }
3382
3383 template <typename A>
3384 bool Parser<A>::altEntryFromSymbol(const macho_nlist<P>& sym)
3385 {
3386 return ( sym.n_desc() & N_ALT_ENTRY );
3387 }
3388
3389
3390 /* Skip over a LEB128 value (signed or unsigned). */
3391 static void
3392 skip_leb128 (const uint8_t ** offset, const uint8_t * end)
3393 {
3394 while (*offset != end && **offset >= 0x80)
3395 (*offset)++;
3396 if (*offset != end)
3397 (*offset)++;
3398 }
3399
3400 /* Read a ULEB128 into a 64-bit word. Return (uint64_t)-1 on overflow
3401 or error. On overflow, skip past the rest of the uleb128. */
3402 static uint64_t
3403 read_uleb128 (const uint8_t ** offset, const uint8_t * end)
3404 {
3405 uint64_t result = 0;
3406 int bit = 0;
3407
3408 do {
3409 uint64_t b;
3410
3411 if (*offset == end)
3412 return (uint64_t) -1;
3413
3414 b = **offset & 0x7f;
3415
3416 if (bit >= 64 || b << bit >> bit != b)
3417 result = (uint64_t) -1;
3418 else
3419 result |= b << bit, bit += 7;
3420 } while (*(*offset)++ >= 0x80);
3421 return result;
3422 }
3423
3424
3425 /* Skip over a DWARF attribute of form FORM. */
3426 template <typename A>
3427 bool Parser<A>::skip_form(const uint8_t ** offset, const uint8_t * end, uint64_t form,
3428 uint8_t addr_size, bool dwarf64)
3429 {
3430 int64_t sz=0;
3431
3432 switch (form)
3433 {
3434 case DW_FORM_addr:
3435 sz = addr_size;
3436 break;
3437
3438 case DW_FORM_block2:
3439 if (end - *offset < 2)
3440 return false;
3441 sz = 2 + A::P::E::get16(*(uint16_t*)offset);
3442 break;
3443
3444 case DW_FORM_block4:
3445 if (end - *offset < 4)
3446 return false;
3447 sz = 2 + A::P::E::get32(*(uint32_t*)offset);
3448 break;
3449
3450 case DW_FORM_data2:
3451 case DW_FORM_ref2:
3452 sz = 2;
3453 break;
3454
3455 case DW_FORM_data4:
3456 case DW_FORM_ref4:
3457 sz = 4;
3458 break;
3459
3460 case DW_FORM_data8:
3461 case DW_FORM_ref8:
3462 sz = 8;
3463 break;
3464
3465 case DW_FORM_string:
3466 while (*offset != end && **offset)
3467 ++*offset;
3468 case DW_FORM_data1:
3469 case DW_FORM_flag:
3470 case DW_FORM_ref1:
3471 sz = 1;
3472 break;
3473
3474 case DW_FORM_block:
3475 sz = read_uleb128 (offset, end);
3476 break;
3477
3478 case DW_FORM_block1:
3479 if (*offset == end)
3480 return false;
3481 sz = 1 + **offset;
3482 break;
3483
3484 case DW_FORM_sdata:
3485 case DW_FORM_udata:
3486 case DW_FORM_ref_udata:
3487 skip_leb128 (offset, end);
3488 return true;
3489
3490 case DW_FORM_strp:
3491 case DW_FORM_ref_addr:
3492 sz = 4;
3493 break;
3494
3495 case DW_FORM_sec_offset:
3496 sz = sizeof(typename A::P::uint_t);
3497 break;
3498
3499 case DW_FORM_exprloc:
3500 sz = read_uleb128 (offset, end);
3501 break;
3502
3503 case DW_FORM_flag_present:
3504 sz = 0;
3505 break;
3506
3507 case DW_FORM_ref_sig8:
3508 sz = 8;
3509 break;
3510
3511 default:
3512 return false;
3513 }
3514 if (end - *offset < sz)
3515 return false;
3516 *offset += sz;
3517 return true;
3518 }
3519
3520
3521 template <typename A>
3522 const char* Parser<A>::getDwarfString(uint64_t form, const uint8_t*& di)
3523 {
3524 uint32_t offset;
3525 const char* dwarfStrings;
3526 const char* result = NULL;
3527 switch (form) {
3528 case DW_FORM_string:
3529 result = (const char*)di;
3530 di += strlen(result) + 1;
3531 break;
3532 case DW_FORM_strp:
3533 offset = E::get32(*((uint32_t*)di));
3534 dwarfStrings = (char*)_file->fileContent() + _file->_dwarfDebugStringSect->offset();
3535 if ( offset < _file->_dwarfDebugStringSect->size() )
3536 result = &dwarfStrings[offset];
3537 else
3538 warning("dwarf DW_FORM_strp (offset=0x%08X) is too big in %s", offset, this->_path);
3539 di += 4;
3540 break;
3541 default:
3542 warning("unknown dwarf string encoding (form=%lld) in %s", form, this->_path);
3543 break;
3544 }
3545 return result;
3546 }
3547
3548 template <typename A>
3549 uint64_t Parser<A>::getDwarfOffset(uint64_t form, const uint8_t*& di, bool dwarf64)
3550 {
3551 if ( form == DW_FORM_sec_offset )
3552 form = (dwarf64 ? DW_FORM_data8 : DW_FORM_data4);
3553 uint64_t result = -1;
3554 switch (form) {
3555 case DW_FORM_data4:
3556 result = A::P::E::get32(*(uint32_t*)di);
3557 di += 4;
3558 break;
3559 case DW_FORM_data8:
3560 result = A::P::E::get64(*(uint64_t*)di);
3561 di += 8;
3562 break;
3563 default:
3564 warning("unknown dwarf DW_FORM_ for DW_AT_stmt_list in %s", this->_path);
3565 }
3566 return result;
3567 }
3568
3569
3570 template <typename A>
3571 struct AtomAndLineInfo {
3572 Atom<A>* atom;
3573 ld::Atom::LineInfo info;
3574 };
3575
3576
3577 // <rdar://problem/5591394> Add support to ld64 for N_FUN stabs when used for symbolic constants
3578 // Returns whether a stabStr belonging to an N_FUN stab represents a
3579 // symbolic constant rather than a function
3580 template <typename A>
3581 bool Parser<A>::isConstFunStabs(const char *stabStr)
3582 {
3583 const char* colon;
3584 // N_FUN can be used for both constants and for functions. In case it's a constant,
3585 // the format of the stabs string is "symname:c=<value>;"
3586 // ':' cannot appear in the symbol name, except if it's an Objective-C method
3587 // (in which case the symbol name starts with + or -, and then it's definitely
3588 // not a constant)
3589 return (stabStr != NULL) && (stabStr[0] != '+') && (stabStr[0] != '-')
3590 && ((colon = strchr(stabStr, ':')) != NULL)
3591 && (colon[1] == 'c') && (colon[2] == '=');
3592 }
3593
3594
3595 template <typename A>
3596 void Parser<A>::parseDebugInfo()
3597 {
3598 // check for dwarf __debug_info section
3599 if ( _file->_dwarfDebugInfoSect == NULL ) {
3600 // if no DWARF debug info, look for stabs
3601 this->parseStabs();
3602 return;
3603 }
3604 if ( _file->_dwarfDebugInfoSect->size() == 0 )
3605 return;
3606
3607 uint64_t stmtList;
3608 const char* tuDir;
3609 const char* tuName;
3610 if ( !read_comp_unit(&tuName, &tuDir, &stmtList) ) {
3611 // if can't parse dwarf, warn and give up
3612 _file->_dwarfTranslationUnitPath = NULL;
3613 warning("can't parse dwarf compilation unit info in %s", _path);
3614 _file->_debugInfoKind = ld::relocatable::File::kDebugInfoNone;
3615 return;
3616 }
3617 if ( (tuName != NULL) && (tuName[0] == '/') ) {
3618 _file->_dwarfTranslationUnitPath = tuName;
3619 }
3620 else if ( (tuDir != NULL) && (tuName != NULL) ) {
3621 asprintf((char**)&(_file->_dwarfTranslationUnitPath), "%s/%s", tuDir, tuName);
3622 }
3623 else if ( tuDir == NULL ) {
3624 _file->_dwarfTranslationUnitPath = tuName;
3625 }
3626 else {
3627 _file->_dwarfTranslationUnitPath = NULL;
3628 }
3629
3630 // add line number info to atoms from dwarf
3631 std::vector<AtomAndLineInfo<A> > entries;
3632 entries.reserve(64);
3633 if ( _file->_debugInfoKind == ld::relocatable::File::kDebugInfoDwarf ) {
3634 // file with just data will have no __debug_line info
3635 if ( (_file->_dwarfDebugLineSect != NULL) && (_file->_dwarfDebugLineSect->size() != 0) ) {
3636 // validate stmt_list
3637 if ( (stmtList != (uint64_t)-1) && (stmtList < _file->_dwarfDebugLineSect->size()) ) {
3638 const uint8_t* debug_line = (uint8_t*)_file->fileContent() + _file->_dwarfDebugLineSect->offset();
3639 struct line_reader_data* lines = line_open(&debug_line[stmtList],
3640 _file->_dwarfDebugLineSect->size() - stmtList, E::little_endian);
3641 struct line_info result;
3642 Atom<A>* curAtom = NULL;
3643 uint32_t curAtomOffset = 0;
3644 uint32_t curAtomAddress = 0;
3645 uint32_t curAtomSize = 0;
3646 std::map<uint32_t,const char*> dwarfIndexToFile;
3647 if ( lines != NULL ) {
3648 while ( line_next(lines, &result, line_stop_pc) ) {
3649 //fprintf(stderr, "curAtom=%p, result.pc=0x%llX, result.line=%llu, result.end_of_sequence=%d,"
3650 // " curAtomAddress=0x%X, curAtomSize=0x%X\n",
3651 // curAtom, result.pc, result.line, result.end_of_sequence, curAtomAddress, curAtomSize);
3652 // work around weird debug line table compiler generates if no functions in __text section
3653 if ( (curAtom == NULL) && (result.pc == 0) && result.end_of_sequence && (result.file == 1))
3654 continue;
3655 // for performance, see if in next pc is in current atom
3656 if ( (curAtom != NULL) && (curAtomAddress <= result.pc) && (result.pc < (curAtomAddress+curAtomSize)) ) {
3657 curAtomOffset = result.pc - curAtomAddress;
3658 }
3659 // or pc at end of current atom
3660 else if ( result.end_of_sequence && (curAtom != NULL) && (result.pc == (curAtomAddress+curAtomSize)) ) {
3661 curAtomOffset = result.pc - curAtomAddress;
3662 }
3663 // or only one function that is a one line function
3664 else if ( result.end_of_sequence && (curAtom == NULL) && (this->findAtomByAddress(0) != NULL) && (result.pc == this->findAtomByAddress(0)->size()) ) {
3665 curAtom = this->findAtomByAddress(0);
3666 curAtomOffset = result.pc - curAtom->objectAddress();
3667 curAtomAddress = curAtom->objectAddress();
3668 curAtomSize = curAtom->size();
3669 }
3670 else {
3671 // do slow look up of atom by address
3672 try {
3673 curAtom = this->findAtomByAddress(result.pc);
3674 }
3675 catch (...) {
3676 // in case of bug in debug info, don't abort link, just limp on
3677 curAtom = NULL;
3678 }
3679 if ( curAtom == NULL )
3680 break; // file has line info but no functions
3681 if ( result.end_of_sequence && (curAtomAddress+curAtomSize < result.pc) ) {
3682 // a one line function can be returned by line_next() as one entry with pc at end of blob
3683 // look for alt atom starting at end of previous atom
3684 uint32_t previousEnd = curAtomAddress+curAtomSize;
3685 Atom<A>* alt = this->findAtomByAddressOrNullIfStub(previousEnd);
3686 if ( alt == NULL )
3687 continue; // ignore spurious debug info for stubs
3688 if ( result.pc <= alt->objectAddress() + alt->size() ) {
3689 curAtom = alt;
3690 curAtomOffset = result.pc - alt->objectAddress();
3691 curAtomAddress = alt->objectAddress();
3692 curAtomSize = alt->size();
3693 }
3694 else {
3695 curAtomOffset = result.pc - curAtom->objectAddress();
3696 curAtomAddress = curAtom->objectAddress();
3697 curAtomSize = curAtom->size();
3698 }
3699 }
3700 else {
3701 curAtomOffset = result.pc - curAtom->objectAddress();
3702 curAtomAddress = curAtom->objectAddress();
3703 curAtomSize = curAtom->size();
3704 }
3705 }
3706 const char* filename;
3707 std::map<uint32_t,const char*>::iterator pos = dwarfIndexToFile.find(result.file);
3708 if ( pos == dwarfIndexToFile.end() ) {
3709 filename = line_file(lines, result.file);
3710 dwarfIndexToFile[result.file] = filename;
3711 }
3712 else {
3713 filename = pos->second;
3714 }
3715 // only record for ~8000 line info records per function
3716 if ( curAtom->roomForMoreLineInfoCount() ) {
3717 AtomAndLineInfo<A> entry;
3718 entry.atom = curAtom;
3719 entry.info.atomOffset = curAtomOffset;
3720 entry.info.fileName = filename;
3721 entry.info.lineNumber = result.line;
3722 //fprintf(stderr, "addr=0x%08llX, line=%lld, file=%s, atom=%s, atom.size=0x%X, end=%d\n",
3723 // result.pc, result.line, filename, curAtom->name(), curAtomSize, result.end_of_sequence);
3724 entries.push_back(entry);
3725 curAtom->incrementLineInfoCount();
3726 }
3727 if ( result.end_of_sequence ) {
3728 curAtom = NULL;
3729 }
3730 }
3731 line_free(lines);
3732 }
3733 }
3734 }
3735 }
3736
3737 // assign line info start offset for each atom
3738 uint8_t* p = _file->_atomsArray;
3739 uint32_t liOffset = 0;
3740 for(int i=_file->_atomsArrayCount; i > 0; --i) {
3741 Atom<A>* atom = (Atom<A>*)p;
3742 atom->_lineInfoStartIndex = liOffset;
3743 liOffset += atom->_lineInfoCount;
3744 atom->_lineInfoCount = 0;
3745 p += sizeof(Atom<A>);
3746 }
3747 assert(liOffset == entries.size());
3748 _file->_lineInfos.resize(liOffset);
3749
3750 // copy each line info for each atom
3751 for (typename std::vector<AtomAndLineInfo<A> >::iterator it = entries.begin(); it != entries.end(); ++it) {
3752 uint32_t slot = it->atom->_lineInfoStartIndex + it->atom->_lineInfoCount;
3753 _file->_lineInfos[slot] = it->info;
3754 it->atom->_lineInfoCount++;
3755 }
3756
3757 // done with temp vector
3758 entries.clear();
3759 }
3760
3761 template <typename A>
3762 void Parser<A>::parseStabs()
3763 {
3764 // scan symbol table for stabs entries
3765 Atom<A>* currentAtom = NULL;
3766 pint_t currentAtomAddress = 0;
3767 enum { start, inBeginEnd, inFun } state = start;
3768 for (uint32_t symbolIndex = 0; symbolIndex < _symbolCount; ++symbolIndex ) {
3769 const macho_nlist<P>& sym = this->symbolFromIndex(symbolIndex);
3770 bool useStab = true;
3771 uint8_t type = sym.n_type();
3772 const char* symString = (sym.n_strx() != 0) ? this->nameFromSymbol(sym) : NULL;
3773 if ( (type & N_STAB) != 0 ) {
3774 _file->_debugInfoKind = (_hasUUID ? ld::relocatable::File::kDebugInfoStabsUUID : ld::relocatable::File::kDebugInfoStabs);
3775 ld::relocatable::File::Stab stab;
3776 stab.atom = NULL;
3777 stab.type = type;
3778 stab.other = sym.n_sect();
3779 stab.desc = sym.n_desc();
3780 stab.value = sym.n_value();
3781 stab.string = NULL;
3782 switch (state) {
3783 case start:
3784 switch (type) {
3785 case N_BNSYM:
3786 // beginning of function block
3787 state = inBeginEnd;
3788 // fall into case to lookup atom by addresss
3789 case N_LCSYM:
3790 case N_STSYM:
3791 currentAtomAddress = sym.n_value();
3792 currentAtom = this->findAtomByAddress(currentAtomAddress);
3793 if ( currentAtom != NULL ) {
3794 stab.atom = currentAtom;
3795 stab.string = symString;
3796 }
3797 else {
3798 fprintf(stderr, "can't find atom for stabs BNSYM at %08llX in %s",
3799 (uint64_t)sym.n_value(), _path);
3800 }
3801 break;
3802 case N_SO:
3803 case N_OSO:
3804 case N_OPT:
3805 case N_LSYM:
3806 case N_RSYM:
3807 case N_PSYM:
3808 case N_AST:
3809 // not associated with an atom, just copy
3810 stab.string = symString;
3811 break;
3812 case N_GSYM:
3813 {
3814 // n_value field is NOT atom address ;-(
3815 // need to find atom by name match
3816 const char* colon = strchr(symString, ':');
3817 if ( colon != NULL ) {
3818 // build underscore leading name
3819 int nameLen = colon - symString;
3820 char symName[nameLen+2];
3821 strlcpy(&symName[1], symString, nameLen+1);
3822 symName[0] = '_';
3823 symName[nameLen+1] = '\0';
3824 currentAtom = this->findAtomByName(symName);
3825 if ( currentAtom != NULL ) {
3826 stab.atom = currentAtom;
3827 stab.string = symString;
3828 }
3829 }
3830 else {
3831 // might be a debug-note without trailing :G()
3832 currentAtom = this->findAtomByName(symString);
3833 if ( currentAtom != NULL ) {
3834 stab.atom = currentAtom;
3835 stab.string = symString;
3836 }
3837 }
3838 if ( stab.atom == NULL ) {
3839 // ld_classic added bogus GSYM stabs for old style dtrace probes
3840 if ( (strncmp(symString, "__dtrace_probe$", 15) != 0) )
3841 warning("can't find atom for N_GSYM stabs %s in %s", symString, _path);
3842 useStab = false;
3843 }
3844 break;
3845 }
3846 case N_FUN:
3847 if ( isConstFunStabs(symString) ) {
3848 // constant not associated with a function
3849 stab.string = symString;
3850 }
3851 else {
3852 // old style stabs without BNSYM
3853 state = inFun;
3854 currentAtomAddress = sym.n_value();
3855 currentAtom = this->findAtomByAddress(currentAtomAddress);
3856 if ( currentAtom != NULL ) {
3857 stab.atom = currentAtom;
3858 stab.string = symString;
3859 }
3860 else {
3861 warning("can't find atom for stabs FUN at %08llX in %s",
3862 (uint64_t)currentAtomAddress, _path);
3863 }
3864 }
3865 break;
3866 case N_SOL:
3867 case N_SLINE:
3868 stab.string = symString;
3869 // old stabs
3870 break;
3871 case N_BINCL:
3872 case N_EINCL:
3873 case N_EXCL:
3874 stab.string = symString;
3875 // -gfull built .o file
3876 break;
3877 default:
3878 warning("unknown stabs type 0x%X in %s", type, _path);
3879 }
3880 break;
3881 case inBeginEnd:
3882 stab.atom = currentAtom;
3883 switch (type) {
3884 case N_ENSYM:
3885 state = start;
3886 currentAtom = NULL;
3887 break;
3888 case N_LCSYM:
3889 case N_STSYM:
3890 {
3891 Atom<A>* nestedAtom = this->findAtomByAddress(sym.n_value());
3892 if ( nestedAtom != NULL ) {
3893 stab.atom = nestedAtom;
3894 stab.string = symString;
3895 }
3896 else {
3897 warning("can't find atom for stabs 0x%X at %08llX in %s",
3898 type, (uint64_t)sym.n_value(), _path);
3899 }
3900 break;
3901 }
3902 case N_LBRAC:
3903 case N_RBRAC:
3904 case N_SLINE:
3905 // adjust value to be offset in atom
3906 stab.value -= currentAtomAddress;
3907 default:
3908 stab.string = symString;
3909 break;
3910 }
3911 break;
3912 case inFun:
3913 switch (type) {
3914 case N_FUN:
3915 if ( isConstFunStabs(symString) ) {
3916 stab.atom = currentAtom;
3917 stab.string = symString;
3918 }
3919 else {
3920 if ( sym.n_sect() != 0 ) {
3921 // found another start stab, must be really old stabs...
3922 currentAtomAddress = sym.n_value();
3923 currentAtom = this->findAtomByAddress(currentAtomAddress);
3924 if ( currentAtom != NULL ) {
3925 stab.atom = currentAtom;
3926 stab.string = symString;
3927 }
3928 else {
3929 warning("can't find atom for stabs FUN at %08llX in %s",
3930 (uint64_t)currentAtomAddress, _path);
3931 }
3932 }
3933 else {
3934 // found ending stab, switch back to start state
3935 stab.string = symString;
3936 stab.atom = currentAtom;
3937 state = start;
3938 currentAtom = NULL;
3939 }
3940 }
3941 break;
3942 case N_LBRAC:
3943 case N_RBRAC:
3944 case N_SLINE:
3945 // adjust value to be offset in atom
3946 stab.value -= currentAtomAddress;
3947 stab.atom = currentAtom;
3948 break;
3949 case N_SO:
3950 stab.string = symString;
3951 state = start;
3952 break;
3953 default:
3954 stab.atom = currentAtom;
3955 stab.string = symString;
3956 break;
3957 }
3958 break;
3959 }
3960 // add to list of stabs for this .o file
3961 if ( useStab )
3962 _file->_stabs.push_back(stab);
3963 }
3964 }
3965 }
3966
3967
3968
3969 // Look at the compilation unit DIE and determine
3970 // its NAME, compilation directory (in COMP_DIR) and its
3971 // line number information offset (in STMT_LIST). NAME and COMP_DIR
3972 // may be NULL (especially COMP_DIR) if they are not in the .o file;
3973 // STMT_LIST will be (uint64_t) -1.
3974 //
3975 // At present this assumes that there's only one compilation unit DIE.
3976 //
3977 template <typename A>
3978 bool Parser<A>::read_comp_unit(const char ** name, const char ** comp_dir,
3979 uint64_t *stmt_list)
3980 {
3981 const uint8_t * debug_info;
3982 const uint8_t * debug_abbrev;
3983 const uint8_t * di;
3984 const uint8_t * next_cu;
3985 const uint8_t * da;
3986 const uint8_t * end;
3987 const uint8_t * enda;
3988 uint64_t sz;
3989 uint16_t vers;
3990 uint64_t abbrev_base;
3991 uint64_t abbrev;
3992 uint8_t address_size;
3993 bool dwarf64;
3994
3995 *name = NULL;
3996 *comp_dir = NULL;
3997 *stmt_list = (uint64_t) -1;
3998
3999 if ( (_file->_dwarfDebugInfoSect == NULL) || (_file->_dwarfDebugAbbrevSect == NULL) )
4000 return false;
4001
4002 if (_file->_dwarfDebugInfoSect->size() < 12)
4003 /* Too small to be a real debug_info section. */
4004 return false;
4005
4006 debug_info = (uint8_t*)_file->fileContent() + _file->_dwarfDebugInfoSect->offset();
4007 debug_abbrev = (uint8_t*)_file->fileContent() + _file->_dwarfDebugAbbrevSect->offset();
4008 next_cu = debug_info;
4009
4010 while ((uint64_t)(next_cu - debug_info) < _file->_dwarfDebugInfoSect->size()) {
4011 di = next_cu;
4012 sz = A::P::E::get32(*(uint32_t*)di);
4013 di += 4;
4014 dwarf64 = sz == 0xffffffff;
4015 if (dwarf64)
4016 sz = A::P::E::get64(*(uint64_t*)di), di += 8;
4017 else if (sz > 0xffffff00)
4018 /* Unknown dwarf format. */
4019 return false;
4020
4021 /* Verify claimed size. */
4022 if (sz + (di - debug_info) > _file->_dwarfDebugInfoSect->size() || sz <= (dwarf64 ? 23 : 11))
4023 return false;
4024
4025 next_cu = di + sz;
4026
4027 vers = A::P::E::get16(*(uint16_t*)di);
4028 if (vers < 2 || vers > 4)
4029 /* DWARF version wrong for this code.
4030 Chances are we could continue anyway, but we don't know for sure. */
4031 return false;
4032 di += 2;
4033
4034 /* Find the debug_abbrev section. */
4035 abbrev_base = dwarf64 ? A::P::E::get64(*(uint64_t*)di) : A::P::E::get32(*(uint32_t*)di);
4036 di += dwarf64 ? 8 : 4;
4037
4038 if (abbrev_base > _file->_dwarfDebugAbbrevSect->size())
4039 return false;
4040 da = debug_abbrev + abbrev_base;
4041 enda = debug_abbrev + _file->_dwarfDebugAbbrevSect->size();
4042
4043 address_size = *di++;
4044
4045 /* Find the abbrev number we're looking for. */
4046 end = di + sz;
4047 abbrev = read_uleb128 (&di, end);
4048 if (abbrev == (uint64_t) -1)
4049 return false;
4050
4051 /* Skip through the debug_abbrev section looking for that abbrev. */
4052 for (;;)
4053 {
4054 uint64_t this_abbrev = read_uleb128 (&da, enda);
4055 uint64_t attr;
4056
4057 if (this_abbrev == abbrev)
4058 /* This is almost always taken. */
4059 break;
4060 skip_leb128 (&da, enda); /* Skip the tag. */
4061 if (da == enda)
4062 return false;
4063 da++; /* Skip the DW_CHILDREN_* value. */
4064
4065 do {
4066 attr = read_uleb128 (&da, enda);
4067 skip_leb128 (&da, enda);
4068 } while (attr != 0 && attr != (uint64_t) -1);
4069 if (attr != 0)
4070 return false;
4071 }
4072
4073 /* Check that the abbrev is one for a DW_TAG_compile_unit. */
4074 if (read_uleb128 (&da, enda) != DW_TAG_compile_unit)
4075 return false;
4076 if (da == enda)
4077 return false;
4078 da++; /* Skip the DW_CHILDREN_* value. */
4079
4080 /* Now, go through the DIE looking for DW_AT_name,
4081 DW_AT_comp_dir, and DW_AT_stmt_list. */
4082 bool skip_to_next_cu = false;
4083 while (!skip_to_next_cu) {
4084
4085 uint64_t attr = read_uleb128 (&da, enda);
4086 uint64_t form = read_uleb128 (&da, enda);
4087
4088 if (attr == (uint64_t) -1)
4089 return false;
4090 else if (attr == 0)
4091 return true;
4092 if (form == DW_FORM_indirect)
4093 form = read_uleb128 (&di, end);
4094
4095 switch (attr) {
4096 case DW_AT_name:
4097 *name = getDwarfString(form, di);
4098 /* Swift object files may contain two CUs: One
4099 describes the Swift code, one is created by the
4100 clang importer. Skip over the CU created by the
4101 clang importer as it may be empty. */
4102 if (std::string(*name) == "<swift-imported-modules>")
4103 skip_to_next_cu = true;
4104 break;
4105 case DW_AT_comp_dir:
4106 *comp_dir = getDwarfString(form, di);
4107 break;
4108 case DW_AT_stmt_list:
4109 *stmt_list = getDwarfOffset(form, di, dwarf64);
4110 break;
4111 default:
4112 if (! skip_form (&di, end, form, address_size, dwarf64))
4113 return false;
4114 }
4115 }
4116 }
4117 return false;
4118 }
4119
4120
4121
4122 template <typename A>
4123 File<A>::~File()
4124 {
4125 free(_sectionsArray);
4126 free(_atomsArray);
4127 }
4128
4129 template <typename A>
4130 const char* File<A>::translationUnitSource() const
4131 {
4132 return _dwarfTranslationUnitPath;
4133 }
4134
4135 template <typename A>
4136 bool File<A>::forEachAtom(ld::File::AtomHandler& handler) const
4137 {
4138 handler.doFile(*this);
4139 uint8_t* p = _atomsArray;
4140 for(int i=_atomsArrayCount; i > 0; --i) {
4141 handler.doAtom(*((Atom<A>*)p));
4142 p += sizeof(Atom<A>);
4143 }
4144 p = _aliasAtomsArray;
4145 for(int i=_aliasAtomsArrayCount; i > 0; --i) {
4146 handler.doAtom(*((AliasAtom*)p));
4147 p += sizeof(AliasAtom);
4148 }
4149
4150 return (_atomsArrayCount != 0) || (_aliasAtomsArrayCount != 0);
4151 }
4152
4153 template <typename A>
4154 const char* Section<A>::makeSegmentName(const macho_section<typename A::P>* sect)
4155 {
4156 // mach-o section record only has room for 16-byte seg/sect names
4157 // so a 16-byte name has no trailing zero
4158 const char* name = sect->segname();
4159 if ( strlen(name) < 16 )
4160 return name;
4161 char* tmp = new char[17];
4162 strlcpy(tmp, name, 17);
4163 return tmp;
4164 }
4165
4166 template <typename A>
4167 const char* Section<A>::makeSectionName(const macho_section<typename A::P>* sect)
4168 {
4169 const char* name = sect->sectname();
4170 if ( strlen(name) < 16 )
4171 return name;
4172
4173 // special case common long section names so we don't have to malloc
4174 if ( strncmp(sect->sectname(), "__objc_classrefs", 16) == 0 )
4175 return "__objc_classrefs";
4176 if ( strncmp(sect->sectname(), "__objc_classlist", 16) == 0 )
4177 return "__objc_classlist";
4178 if ( strncmp(sect->sectname(), "__objc_nlclslist", 16) == 0 )
4179 return "__objc_nlclslist";
4180 if ( strncmp(sect->sectname(), "__objc_nlcatlist", 16) == 0 )
4181 return "__objc_nlcatlist";
4182 if ( strncmp(sect->sectname(), "__objc_protolist", 16) == 0 )
4183 return "__objc_protolist";
4184 if ( strncmp(sect->sectname(), "__objc_protorefs", 16) == 0 )
4185 return "__objc_protorefs";
4186 if ( strncmp(sect->sectname(), "__objc_superrefs", 16) == 0 )
4187 return "__objc_superrefs";
4188 if ( strncmp(sect->sectname(), "__objc_imageinfo", 16) == 0 )
4189 return "__objc_imageinfo";
4190 if ( strncmp(sect->sectname(), "__objc_stringobj", 16) == 0 )
4191 return "__objc_stringobj";
4192 if ( strncmp(sect->sectname(), "__gcc_except_tab", 16) == 0 )
4193 return "__gcc_except_tab";
4194
4195 char* tmp = new char[17];
4196 strlcpy(tmp, name, 17);
4197 return tmp;
4198 }
4199
4200 template <typename A>
4201 bool Section<A>::readable(const macho_section<typename A::P>* sect)
4202 {
4203 return true;
4204 }
4205
4206 template <typename A>
4207 bool Section<A>::writable(const macho_section<typename A::P>* sect)
4208 {
4209 // mach-o .o files do not contain segment permissions
4210 // we just know TEXT is special
4211 return ( strcmp(sect->segname(), "__TEXT") != 0 );
4212 }
4213
4214 template <typename A>
4215 bool Section<A>::exectuable(const macho_section<typename A::P>* sect)
4216 {
4217 // mach-o .o files do not contain segment permissions
4218 // we just know TEXT is special
4219 return ( strcmp(sect->segname(), "__TEXT") == 0 );
4220 }
4221
4222
4223 template <typename A>
4224 ld::Section::Type Section<A>::sectionType(const macho_section<typename A::P>* sect)
4225 {
4226 switch ( sect->flags() & SECTION_TYPE ) {
4227 case S_ZEROFILL:
4228 return ld::Section::typeZeroFill;
4229 case S_CSTRING_LITERALS:
4230 if ( (strcmp(sect->sectname(), "__cstring") == 0) && (strcmp(sect->segname(), "__TEXT") == 0) )
4231 return ld::Section::typeCString;
4232 else
4233 return ld::Section::typeNonStdCString;
4234 case S_4BYTE_LITERALS:
4235 return ld::Section::typeLiteral4;
4236 case S_8BYTE_LITERALS:
4237 return ld::Section::typeLiteral8;
4238 case S_LITERAL_POINTERS:
4239 return ld::Section::typeCStringPointer;
4240 case S_NON_LAZY_SYMBOL_POINTERS:
4241 return ld::Section::typeNonLazyPointer;
4242 case S_LAZY_SYMBOL_POINTERS:
4243 return ld::Section::typeLazyPointer;
4244 case S_SYMBOL_STUBS:
4245 return ld::Section::typeStub;
4246 case S_MOD_INIT_FUNC_POINTERS:
4247 return ld::Section::typeInitializerPointers;
4248 case S_MOD_TERM_FUNC_POINTERS:
4249 return ld::Section::typeTerminatorPointers;
4250 case S_INTERPOSING:
4251 return ld::Section::typeUnclassified;
4252 case S_16BYTE_LITERALS:
4253 return ld::Section::typeLiteral16;
4254 case S_REGULAR:
4255 case S_COALESCED:
4256 if ( sect->flags() & S_ATTR_PURE_INSTRUCTIONS ) {
4257 return ld::Section::typeCode;
4258 }
4259 else if ( strcmp(sect->segname(), "__TEXT") == 0 ) {
4260 if ( strcmp(sect->sectname(), "__eh_frame") == 0 )
4261 return ld::Section::typeCFI;
4262 else if ( strcmp(sect->sectname(), "__ustring") == 0 )
4263 return ld::Section::typeUTF16Strings;
4264 else if ( strcmp(sect->sectname(), "__textcoal_nt") == 0 )
4265 return ld::Section::typeCode;
4266 else if ( strcmp(sect->sectname(), "__StaticInit") == 0 )
4267 return ld::Section::typeCode;
4268 else if ( strcmp(sect->sectname(), "__constructor") == 0 )
4269 return ld::Section::typeInitializerPointers;
4270 }
4271 else if ( strcmp(sect->segname(), "__DATA") == 0 ) {
4272 if ( strcmp(sect->sectname(), "__cfstring") == 0 )
4273 return ld::Section::typeCFString;
4274 else if ( strcmp(sect->sectname(), "__dyld") == 0 )
4275 return ld::Section::typeDyldInfo;
4276 else if ( strcmp(sect->sectname(), "__program_vars") == 0 )
4277 return ld::Section::typeDyldInfo;
4278 else if ( strncmp(sect->sectname(), "__objc_classrefs", 16) == 0 )
4279 return ld::Section::typeObjCClassRefs;
4280 else if ( strcmp(sect->sectname(), "__objc_catlist") == 0 )
4281 return ld::Section::typeObjC2CategoryList;
4282 }
4283 else if ( strcmp(sect->segname(), "__OBJC") == 0 ) {
4284 if ( strcmp(sect->sectname(), "__class") == 0 )
4285 return ld::Section::typeObjC1Classes;
4286 }
4287 break;
4288 case S_THREAD_LOCAL_REGULAR:
4289 return ld::Section::typeTLVInitialValues;
4290 case S_THREAD_LOCAL_ZEROFILL:
4291 return ld::Section::typeTLVZeroFill;
4292 case S_THREAD_LOCAL_VARIABLES:
4293 return ld::Section::typeTLVDefs;
4294 case S_THREAD_LOCAL_VARIABLE_POINTERS:
4295 return ld::Section::typeTLVPointers;
4296 case S_THREAD_LOCAL_INIT_FUNCTION_POINTERS:
4297 return ld::Section::typeTLVInitializerPointers;
4298 }
4299 return ld::Section::typeUnclassified;
4300 }
4301
4302
4303 template <typename A>
4304 Atom<A>* Section<A>::findContentAtomByAddress(pint_t addr, class Atom<A>* start, class Atom<A>* end)
4305 {
4306 // do a binary search of atom array
4307 uint32_t atomCount = end - start;
4308 Atom<A>* base = start;
4309 for (uint32_t n = atomCount; n > 0; n /= 2) {
4310 Atom<A>* pivot = &base[n/2];
4311 pint_t atomStartAddr = pivot->_objAddress;
4312 pint_t atomEndAddr = atomStartAddr + pivot->_size;
4313 if ( atomStartAddr <= addr ) {
4314 // address in normal atom
4315 if (addr < atomEndAddr)
4316 return pivot;
4317 // address in "end" label (but not in alias)
4318 if ( (pivot->_size == 0) && (addr == atomEndAddr) && !pivot->isAlias() )
4319 return pivot;
4320 }
4321 if ( addr >= atomEndAddr ) {
4322 // key > pivot
4323 // move base to atom after pivot
4324 base = &pivot[1];
4325 --n;
4326 }
4327 else {
4328 // key < pivot
4329 // keep same base
4330 }
4331 }
4332 return NULL;
4333 }
4334
4335 template <typename A>
4336 ld::Atom::Alignment Section<A>::alignmentForAddress(pint_t addr)
4337 {
4338 const uint32_t sectionAlignment = this->_machOSection->align();
4339 uint32_t modulus = (addr % (1 << sectionAlignment));
4340 if ( modulus > 0xFFFF )
4341 warning("alignment for symbol at address 0x%08llX in %s exceeds 2^16", (uint64_t)addr, this->file().path());
4342 return ld::Atom::Alignment(sectionAlignment, modulus);
4343 }
4344
4345 template <typename A>
4346 uint32_t Section<A>::sectionNum(class Parser<A>& parser) const
4347 {
4348 if ( _machOSection == NULL )
4349 return 0;
4350 else
4351 return 1 + (this->_machOSection - parser.firstMachOSection());
4352 }
4353
4354 // arm does not have zero cost exceptions
4355 template <>
4356 uint32_t CFISection<arm>::cfiCount(Parser<arm>& parser)
4357 {
4358 if ( parser.armUsesZeroCostExceptions() ) {
4359 // create ObjectAddressSpace object for use by libunwind
4360 OAS oas(*this, (uint8_t*)this->file().fileContent()+this->_machOSection->offset());
4361 return libunwind::CFI_Parser<OAS>::getCFICount(oas,
4362 this->_machOSection->addr(), this->_machOSection->size());
4363 }
4364 return 0;
4365 }
4366
4367 template <typename A>
4368 uint32_t CFISection<A>::cfiCount(Parser<A>& parser)
4369 {
4370 // create ObjectAddressSpace object for use by libunwind
4371 OAS oas(*this, (uint8_t*)this->file().fileContent()+this->_machOSection->offset());
4372 return libunwind::CFI_Parser<OAS>::getCFICount(oas,
4373 this->_machOSection->addr(), this->_machOSection->size());
4374 }
4375
4376 template <typename A>
4377 void CFISection<A>::warnFunc(void* ref, uint64_t funcAddr, const char* msg)
4378 {
4379 Parser<A>* parser = (Parser<A>*)ref;
4380 if ( ! parser->warnUnwindConversionProblems() )
4381 return;
4382 if ( funcAddr != CFI_INVALID_ADDRESS ) {
4383 // atoms are not constructed yet, so scan symbol table for labels
4384 const char* name = parser->scanSymbolTableForAddress(funcAddr);
4385 warning("could not create compact unwind for %s: %s", name, msg);
4386 }
4387 else {
4388 warning("could not create compact unwind: %s", msg);
4389 }
4390 }
4391
4392 template <>
4393 bool CFISection<x86_64>::needsRelocating()
4394 {
4395 return true;
4396 }
4397
4398 template <>
4399 bool CFISection<arm64>::needsRelocating()
4400 {
4401 return true;
4402 }
4403
4404
4405 template <typename A>
4406 bool CFISection<A>::needsRelocating()
4407 {
4408 return false;
4409 }
4410
4411 template <>
4412 void CFISection<x86_64>::cfiParse(class Parser<x86_64>& parser, uint8_t* buffer,
4413 libunwind::CFI_Atom_Info<CFISection<x86_64>::OAS>::CFI_Atom_Info cfiArray[],
4414 uint32_t& count, const pint_t cuStarts[], uint32_t cuCount)
4415 {
4416 const uint32_t sectionSize = this->_machOSection->size();
4417 // copy __eh_frame data to buffer
4418 memcpy(buffer, file().fileContent() + this->_machOSection->offset(), sectionSize);
4419
4420 // and apply relocations
4421 const macho_relocation_info<P>* relocs = (macho_relocation_info<P>*)(file().fileContent() + this->_machOSection->reloff());
4422 const macho_relocation_info<P>* relocsEnd = &relocs[this->_machOSection->nreloc()];
4423 for (const macho_relocation_info<P>* reloc = relocs; reloc < relocsEnd; ++reloc) {
4424 uint64_t value = 0;
4425 switch ( reloc->r_type() ) {
4426 case X86_64_RELOC_SUBTRACTOR:
4427 value = 0 - parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4428 ++reloc;
4429 if ( reloc->r_extern() )
4430 value += parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4431 break;
4432 case X86_64_RELOC_UNSIGNED:
4433 value = parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4434 break;
4435 case X86_64_RELOC_GOT:
4436 // this is used for the reference to the personality function in CIEs
4437 // store the symbol number of the personality function for later use as a Fixup
4438 value = reloc->r_symbolnum();
4439 break;
4440 default:
4441 fprintf(stderr, "CFISection::cfiParse() unexpected relocation type at r_address=0x%08X\n", reloc->r_address());
4442 break;
4443 }
4444 if ( reloc->r_address() > sectionSize )
4445 throwf("malformed __eh_frame relocation, offset (0x%08X) is beyond end of section,", reloc->r_address());
4446 uint64_t* p64;
4447 uint32_t* p32;
4448 switch ( reloc->r_length() ) {
4449 case 3:
4450 p64 = (uint64_t*)&buffer[reloc->r_address()];
4451 E::set64(*p64, value + E::get64(*p64));
4452 break;
4453 case 2:
4454 p32 = (uint32_t*)&buffer[reloc->r_address()];
4455 E::set32(*p32, value + E::get32(*p32));
4456 break;
4457 default:
4458 fprintf(stderr, "CFISection::cfiParse() unexpected relocation size at r_address=0x%08X\n", reloc->r_address());
4459 break;
4460 }
4461 }
4462
4463 // create ObjectAddressSpace object for use by libunwind
4464 OAS oas(*this, buffer);
4465
4466 // use libuwind to parse __eh_frame data into array of CFI_Atom_Info
4467 const char* msg;
4468 msg = libunwind::DwarfInstructions<OAS, libunwind::Registers_x86_64>::parseCFIs(
4469 oas, this->_machOSection->addr(), this->_machOSection->size(),
4470 cuStarts, cuCount, parser.keepDwarfUnwind(), parser.forceDwarfConversion(), parser.neverConvertDwarf(),
4471 cfiArray, count, (void*)&parser, warnFunc);
4472 if ( msg != NULL )
4473 throwf("malformed __eh_frame section: %s", msg);
4474 }
4475
4476 template <>
4477 void CFISection<x86>::cfiParse(class Parser<x86>& parser, uint8_t* buffer,
4478 libunwind::CFI_Atom_Info<CFISection<x86>::OAS>::CFI_Atom_Info cfiArray[],
4479 uint32_t& count, const pint_t cuStarts[], uint32_t cuCount)
4480 {
4481 // create ObjectAddressSpace object for use by libunwind
4482 OAS oas(*this, (uint8_t*)this->file().fileContent()+this->_machOSection->offset());
4483
4484 // use libuwind to parse __eh_frame data into array of CFI_Atom_Info
4485 const char* msg;
4486 msg = libunwind::DwarfInstructions<OAS, libunwind::Registers_x86>::parseCFIs(
4487 oas, this->_machOSection->addr(), this->_machOSection->size(),
4488 cuStarts, cuCount, parser.keepDwarfUnwind(), parser.forceDwarfConversion(), parser.neverConvertDwarf(),
4489 cfiArray, count, (void*)&parser, warnFunc);
4490 if ( msg != NULL )
4491 throwf("malformed __eh_frame section: %s", msg);
4492 }
4493
4494
4495
4496
4497 template <>
4498 void CFISection<arm>::cfiParse(class Parser<arm>& parser, uint8_t* buffer,
4499 libunwind::CFI_Atom_Info<CFISection<arm>::OAS>::CFI_Atom_Info cfiArray[],
4500 uint32_t& count, const pint_t cuStarts[], uint32_t cuCount)
4501 {
4502 if ( !parser.armUsesZeroCostExceptions() ) {
4503 // most arm do not use zero cost exceptions
4504 assert(count == 0);
4505 return;
4506 }
4507 // create ObjectAddressSpace object for use by libunwind
4508 OAS oas(*this, (uint8_t*)this->file().fileContent()+this->_machOSection->offset());
4509
4510 // use libuwind to parse __eh_frame data into array of CFI_Atom_Info
4511 const char* msg;
4512 msg = libunwind::DwarfInstructions<OAS, libunwind::Registers_arm>::parseCFIs(
4513 oas, this->_machOSection->addr(), this->_machOSection->size(),
4514 cuStarts, cuCount, parser.keepDwarfUnwind(), parser.forceDwarfConversion(), parser.neverConvertDwarf(),
4515 cfiArray, count, (void*)&parser, warnFunc);
4516 if ( msg != NULL )
4517 throwf("malformed __eh_frame section: %s", msg);
4518 }
4519
4520
4521
4522
4523 template <>
4524 void CFISection<arm64>::cfiParse(class Parser<arm64>& parser, uint8_t* buffer,
4525 libunwind::CFI_Atom_Info<CFISection<arm64>::OAS>::CFI_Atom_Info cfiArray[],
4526 uint32_t& count, const pint_t cuStarts[], uint32_t cuCount)
4527 {
4528 // copy __eh_frame data to buffer
4529 const uint32_t sectionSize = this->_machOSection->size();
4530 memcpy(buffer, file().fileContent() + this->_machOSection->offset(), sectionSize);
4531
4532 // and apply relocations
4533 const macho_relocation_info<P>* relocs = (macho_relocation_info<P>*)(file().fileContent() + this->_machOSection->reloff());
4534 const macho_relocation_info<P>* relocsEnd = &relocs[this->_machOSection->nreloc()];
4535 for (const macho_relocation_info<P>* reloc = relocs; reloc < relocsEnd; ++reloc) {
4536 uint64_t* p64 = (uint64_t*)&buffer[reloc->r_address()];
4537 uint32_t* p32 = (uint32_t*)&buffer[reloc->r_address()];
4538 uint32_t addend32 = E::get32(*p32);
4539 uint64_t addend64 = E::get64(*p64);
4540 uint64_t value = 0;
4541 switch ( reloc->r_type() ) {
4542 case ARM64_RELOC_SUBTRACTOR:
4543 value = 0 - parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4544 ++reloc;
4545 if ( reloc->r_extern() )
4546 value += parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4547 break;
4548 case ARM64_RELOC_UNSIGNED:
4549 value = parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4550 break;
4551 case ARM64_RELOC_POINTER_TO_GOT:
4552 // this is used for the reference to the personality function in CIEs
4553 // store the symbol number of the personality function for later use as a Fixup
4554 value = reloc->r_symbolnum();
4555 addend32 = 0;
4556 addend64 = 0;
4557 break;
4558 default:
4559 fprintf(stderr, "CFISection::cfiParse() unexpected relocation type at r_address=0x%08X\n", reloc->r_address());
4560 break;
4561 }
4562 if ( reloc->r_address() > sectionSize )
4563 throwf("malformed __eh_frame relocation, offset (0x%08X) is beyond end of section,", reloc->r_address());
4564 switch ( reloc->r_length() ) {
4565 case 3:
4566 E::set64(*p64, value + addend64);
4567 break;
4568 case 2:
4569 E::set32(*p32, value + addend32);
4570 break;
4571 default:
4572 fprintf(stderr, "CFISection::cfiParse() unexpected relocation size at r_address=0x%08X\n", reloc->r_address());
4573 break;
4574 }
4575 }
4576
4577
4578 // create ObjectAddressSpace object for use by libunwind
4579 OAS oas(*this, buffer);
4580
4581 // use libuwind to parse __eh_frame data into array of CFI_Atom_Info
4582 const char* msg;
4583 msg = libunwind::DwarfInstructions<OAS, libunwind::Registers_arm64>::parseCFIs(
4584 oas, this->_machOSection->addr(), this->_machOSection->size(),
4585 cuStarts, cuCount, parser.keepDwarfUnwind(), parser.forceDwarfConversion(), parser.neverConvertDwarf(),
4586 cfiArray, count, (void*)&parser, warnFunc);
4587 if ( msg != NULL )
4588 throwf("malformed __eh_frame section: %s", msg);
4589 }
4590
4591
4592 template <typename A>
4593 uint32_t CFISection<A>::computeAtomCount(class Parser<A>& parser,
4594 struct Parser<A>::LabelAndCFIBreakIterator& it,
4595 const struct Parser<A>::CFI_CU_InfoArrays& cfis)
4596 {
4597 return cfis.cfiCount;
4598 }
4599
4600
4601
4602 template <typename A>
4603 uint32_t CFISection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
4604 struct Parser<A>::LabelAndCFIBreakIterator& it,
4605 const struct Parser<A>::CFI_CU_InfoArrays& cfis)
4606 {
4607 this->_beginAtoms = (Atom<A>*)p;
4608 // walk CFI_Atom_Info array and create atom for each entry
4609 const CFI_Atom_Info* start = &cfis.cfiArray[0];
4610 const CFI_Atom_Info* end = &cfis.cfiArray[cfis.cfiCount];
4611 for(const CFI_Atom_Info* a=start; a < end; ++a) {
4612 Atom<A>* space = (Atom<A>*)p;
4613 new (space) Atom<A>(*this, (a->isCIE ? "CIE" : "FDE"), a->address, a->size,
4614 ld::Atom::definitionRegular, ld::Atom::combineNever, ld::Atom::scopeTranslationUnit,
4615 ld::Atom::typeCFI, ld::Atom::symbolTableNotInFinalLinkedImages,
4616 false, false, false, ld::Atom::Alignment(0));
4617 p += sizeof(Atom<A>);
4618 }
4619 this->_endAtoms = (Atom<A>*)p;
4620 return cfis.cfiCount;
4621 }
4622
4623
4624 template <> bool CFISection<x86_64>::bigEndian() { return false; }
4625 template <> bool CFISection<x86>::bigEndian() { return false; }
4626 template <> bool CFISection<arm>::bigEndian() { return false; }
4627 template <> bool CFISection<arm64>::bigEndian() { return false; }
4628
4629 template <>
4630 void CFISection<x86_64>::addCiePersonalityFixups(class Parser<x86_64>& parser, const CFI_Atom_Info* cieInfo)
4631 {
4632 uint8_t personalityEncoding = cieInfo->u.cieInfo.personality.encodingOfTargetAddress;
4633 if ( personalityEncoding == 0x9B ) {
4634 // compiler always produces X86_64_RELOC_GOT with addend of 4 to personality function
4635 // CFISection<x86_64>::cfiParse() set targetAddress to be symbolIndex + 4 + addressInCIE
4636 uint32_t symbolIndex = cieInfo->u.cieInfo.personality.targetAddress - 4
4637 - cieInfo->address - cieInfo->u.cieInfo.personality.offsetInCFI;
4638 const macho_nlist<P>& sym = parser.symbolFromIndex(symbolIndex);
4639 const char* personalityName = parser.nameFromSymbol(sym);
4640
4641 Atom<x86_64>* cieAtom = this->findAtomByAddress(cieInfo->address);
4642 Parser<x86_64>::SourceLocation src(cieAtom, cieInfo->u.cieInfo.personality.offsetInCFI);
4643 parser.addFixup(src, ld::Fixup::k1of3, ld::Fixup::kindSetTargetAddress, false, personalityName);
4644 parser.addFixup(src, ld::Fixup::k2of3, ld::Fixup::kindAddAddend, 4);
4645 parser.addFixup(src, ld::Fixup::k3of3, ld::Fixup::kindStoreX86PCRel32GOT);
4646 }
4647 else if ( personalityEncoding != 0 ) {
4648 throwf("unsupported address encoding (%02X) of personality function in CIE",
4649 personalityEncoding);
4650 }
4651 }
4652
4653 template <>
4654 void CFISection<x86>::addCiePersonalityFixups(class Parser<x86>& parser, const CFI_Atom_Info* cieInfo)
4655 {
4656 uint8_t personalityEncoding = cieInfo->u.cieInfo.personality.encodingOfTargetAddress;
4657 if ( (personalityEncoding == 0x9B) || (personalityEncoding == 0x90) ) {
4658 uint32_t offsetInCFI = cieInfo->u.cieInfo.personality.offsetInCFI;
4659 uint32_t nlpAddr = cieInfo->u.cieInfo.personality.targetAddress;
4660 Atom<x86>* cieAtom = this->findAtomByAddress(cieInfo->address);
4661 Atom<x86>* nlpAtom = parser.findAtomByAddress(nlpAddr);
4662 assert(nlpAtom->contentType() == ld::Atom::typeNonLazyPointer);
4663 Parser<x86>::SourceLocation src(cieAtom, cieInfo->u.cieInfo.personality.offsetInCFI);
4664
4665 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, nlpAtom);
4666 parser.addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, cieAtom);
4667 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, offsetInCFI);
4668 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian32);
4669 }
4670 else if ( personalityEncoding != 0 ) {
4671 throwf("unsupported address encoding (%02X) of personality function in CIE", personalityEncoding);
4672 }
4673 }
4674
4675 #if SUPPORT_ARCH_arm64
4676 template <>
4677 void CFISection<arm64>::addCiePersonalityFixups(class Parser<arm64>& parser, const CFI_Atom_Info* cieInfo)
4678 {
4679 uint8_t personalityEncoding = cieInfo->u.cieInfo.personality.encodingOfTargetAddress;
4680 if ( personalityEncoding == 0x9B ) {
4681 // compiler always produces ARM64_RELOC_GOT r_pcrel=1 to personality function
4682 // CFISection<arm64>::cfiParse() set targetAddress to be symbolIndex + addressInCIE
4683 uint32_t symbolIndex = cieInfo->u.cieInfo.personality.targetAddress
4684 - cieInfo->address - cieInfo->u.cieInfo.personality.offsetInCFI;
4685 const macho_nlist<P>& sym = parser.symbolFromIndex(symbolIndex);
4686 const char* personalityName = parser.nameFromSymbol(sym);
4687
4688 Atom<arm64>* cieAtom = this->findAtomByAddress(cieInfo->address);
4689 Parser<arm64>::SourceLocation src(cieAtom, cieInfo->u.cieInfo.personality.offsetInCFI);
4690 parser.addFixup(src, ld::Fixup::k1of2, ld::Fixup::kindSetTargetAddress, false, personalityName);
4691 parser.addFixup(src, ld::Fixup::k2of2, ld::Fixup::kindStoreARM64PCRelToGOT);
4692 }
4693 else if ( personalityEncoding != 0 ) {
4694 throwf("unsupported address encoding (%02X) of personality function in CIE",
4695 personalityEncoding);
4696 }
4697 }
4698 #endif
4699
4700
4701 template <>
4702 void CFISection<arm>::addCiePersonalityFixups(class Parser<arm>& parser, const CFI_Atom_Info* cieInfo)
4703 {
4704 uint8_t personalityEncoding = cieInfo->u.cieInfo.personality.encodingOfTargetAddress;
4705 if ( (personalityEncoding == 0x9B) || (personalityEncoding == 0x90) ) {
4706 uint32_t offsetInCFI = cieInfo->u.cieInfo.personality.offsetInCFI;
4707 uint32_t nlpAddr = cieInfo->u.cieInfo.personality.targetAddress;
4708 Atom<arm>* cieAtom = this->findAtomByAddress(cieInfo->address);
4709 Atom<arm>* nlpAtom = parser.findAtomByAddress(nlpAddr);
4710 assert(nlpAtom->contentType() == ld::Atom::typeNonLazyPointer);
4711 Parser<arm>::SourceLocation src(cieAtom, cieInfo->u.cieInfo.personality.offsetInCFI);
4712
4713 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, nlpAtom);
4714 parser.addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, cieAtom);
4715 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, offsetInCFI);
4716 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian32);
4717 }
4718 else if ( personalityEncoding != 0 ) {
4719 throwf("unsupported address encoding (%02X) of personality function in CIE", personalityEncoding);
4720 }
4721 }
4722
4723
4724
4725 template <typename A>
4726 void CFISection<A>::addCiePersonalityFixups(class Parser<A>& parser, const CFI_Atom_Info* cieInfo)
4727 {
4728 assert(0 && "addCiePersonalityFixups() not implemented for arch");
4729 }
4730
4731 template <typename A>
4732 void CFISection<A>::makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays& cfis)
4733 {
4734 ld::Fixup::Kind store32 = bigEndian() ? ld::Fixup::kindStoreBigEndian32 : ld::Fixup::kindStoreLittleEndian32;
4735 ld::Fixup::Kind store64 = bigEndian() ? ld::Fixup::kindStoreBigEndian64 : ld::Fixup::kindStoreLittleEndian64;
4736
4737 // add all references for FDEs, including implicit group references
4738 const CFI_Atom_Info* end = &cfis.cfiArray[cfis.cfiCount];
4739 for(const CFI_Atom_Info* p = &cfis.cfiArray[0]; p < end; ++p) {
4740 if ( p->isCIE ) {
4741 // add reference to personality function if used
4742 if ( p->u.cieInfo.personality.targetAddress != CFI_INVALID_ADDRESS ) {
4743 this->addCiePersonalityFixups(parser, p);
4744 }
4745 }
4746 else {
4747 // find FDE Atom
4748 Atom<A>* fdeAtom = this->findAtomByAddress(p->address);
4749 // find function Atom
4750 Atom<A>* functionAtom = parser.findAtomByAddress(p->u.fdeInfo.function.targetAddress);
4751 // find CIE Atom
4752 Atom<A>* cieAtom = this->findAtomByAddress(p->u.fdeInfo.cie.targetAddress);
4753 // find LSDA Atom
4754 Atom<A>* lsdaAtom = NULL;
4755 if ( p->u.fdeInfo.lsda.targetAddress != CFI_INVALID_ADDRESS ) {
4756 lsdaAtom = parser.findAtomByAddress(p->u.fdeInfo.lsda.targetAddress);
4757 }
4758 // add reference from FDE to CIE (always 32-bit pc-rel)
4759 typename Parser<A>::SourceLocation fdeToCieSrc(fdeAtom, p->u.fdeInfo.cie.offsetInCFI);
4760 parser.addFixup(fdeToCieSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, fdeAtom);
4761 parser.addFixup(fdeToCieSrc, ld::Fixup::k2of4, ld::Fixup::kindAddAddend, p->u.fdeInfo.cie.offsetInCFI);
4762 parser.addFixup(fdeToCieSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, cieAtom);
4763 parser.addFixup(fdeToCieSrc, ld::Fixup::k4of4, store32, cieAtom);
4764
4765 // add reference from FDE to function
4766 typename Parser<A>::SourceLocation fdeToFuncSrc(fdeAtom, p->u.fdeInfo.function.offsetInCFI);
4767 switch (p->u.fdeInfo.function.encodingOfTargetAddress) {
4768 case DW_EH_PE_pcrel|DW_EH_PE_ptr:
4769 if ( sizeof(typename A::P::uint_t) == 8 ) {
4770 parser.addFixup(fdeToFuncSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, functionAtom);
4771 parser.addFixup(fdeToFuncSrc, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, fdeAtom);
4772 parser.addFixup(fdeToFuncSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, p->u.fdeInfo.function.offsetInCFI);
4773 parser.addFixup(fdeToFuncSrc, ld::Fixup::k4of4, store64);
4774 break;
4775 }
4776 // else fall into 32-bit case
4777 case DW_EH_PE_pcrel|DW_EH_PE_sdata4:
4778 parser.addFixup(fdeToFuncSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, functionAtom);
4779 parser.addFixup(fdeToFuncSrc, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, fdeAtom);
4780 parser.addFixup(fdeToFuncSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, p->u.fdeInfo.function.offsetInCFI);
4781 parser.addFixup(fdeToFuncSrc, ld::Fixup::k4of4, store32);
4782 break;
4783 default:
4784 throw "unsupported encoding in FDE of pointer to function";
4785 }
4786
4787 // add reference from FDE to LSDA
4788 typename Parser<A>::SourceLocation fdeToLsdaSrc(fdeAtom, p->u.fdeInfo.lsda.offsetInCFI);
4789 if ( lsdaAtom != NULL ) {
4790 switch (p->u.fdeInfo.lsda.encodingOfTargetAddress) {
4791 case DW_EH_PE_pcrel|DW_EH_PE_ptr:
4792 if ( sizeof(typename A::P::uint_t) == 8 ) {
4793 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, lsdaAtom);
4794 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, fdeAtom);
4795 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, p->u.fdeInfo.lsda.offsetInCFI);
4796 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k4of4, store64);
4797 break;
4798 }
4799 // else fall into 32-bit case
4800 case DW_EH_PE_pcrel|DW_EH_PE_sdata4:
4801 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, lsdaAtom);
4802 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, fdeAtom);
4803 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, p->u.fdeInfo.lsda.offsetInCFI);
4804 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k4of4, store32);
4805 break;
4806 default:
4807 throw "unsupported encoding in FDE of pointer to LSDA";
4808 }
4809 }
4810
4811 // FDE is in group lead by function atom
4812 typename Parser<A>::SourceLocation fdeSrc(functionAtom,0);
4813 parser.addFixup(fdeSrc, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateFDE, fdeAtom);
4814
4815 // LSDA is in group lead by function atom
4816 if ( lsdaAtom != NULL ) {
4817 parser.addFixup(fdeSrc, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateLSDA, lsdaAtom);
4818 }
4819 }
4820 }
4821 }
4822
4823
4824
4825
4826 template <typename A>
4827 const void* CFISection<A>::OAS::mappedAddress(pint_t addr)
4828 {
4829 if ( (_ehFrameStartAddr <= addr) && (addr < _ehFrameEndAddr) )
4830 return &_ehFrameContent[addr-_ehFrameStartAddr];
4831 else {
4832 // requested bytes are not in __eh_frame section
4833 // this can occur when examining the instruction bytes in the __text
4834 File<A>& file = _ehFrameSection.file();
4835 for (uint32_t i=0; i < file._sectionsArrayCount; ++i ) {
4836 const macho_section<typename A::P>* sect = file._sectionsArray[i]->machoSection();
4837 // TentativeDefinitionSection and AbsoluteSymbolSection have no mach-o section
4838 if ( sect != NULL ) {
4839 if ( (sect->addr() <= addr) && (addr < (sect->addr()+sect->size())) ) {
4840 return file.fileContent() + sect->offset() + addr - sect->addr();
4841 }
4842 }
4843 }
4844 throwf("__eh_frame parsing problem. Can't find target of reference to address 0x%08llX", (uint64_t)addr);
4845 }
4846 }
4847
4848
4849 template <typename A>
4850 uint64_t CFISection<A>::OAS::getULEB128(pint_t& logicalAddr, pint_t end)
4851 {
4852 uintptr_t size = (end - logicalAddr);
4853 libunwind::LocalAddressSpace::pint_t laddr = (libunwind::LocalAddressSpace::pint_t)mappedAddress(logicalAddr);
4854 libunwind::LocalAddressSpace::pint_t sladdr = laddr;
4855 uint64_t result = libunwind::LocalAddressSpace::getULEB128(laddr, laddr+size);
4856 logicalAddr += (laddr-sladdr);
4857 return result;
4858 }
4859
4860 template <typename A>
4861 int64_t CFISection<A>::OAS::getSLEB128(pint_t& logicalAddr, pint_t end)
4862 {
4863 uintptr_t size = (end - logicalAddr);
4864 libunwind::LocalAddressSpace::pint_t laddr = (libunwind::LocalAddressSpace::pint_t)mappedAddress(logicalAddr);
4865 libunwind::LocalAddressSpace::pint_t sladdr = laddr;
4866 int64_t result = libunwind::LocalAddressSpace::getSLEB128(laddr, laddr+size);
4867 logicalAddr += (laddr-sladdr);
4868 return result;
4869 }
4870
4871 template <typename A>
4872 typename A::P::uint_t CFISection<A>::OAS::getEncodedP(pint_t& addr, pint_t end, uint8_t encoding)
4873 {
4874 pint_t startAddr = addr;
4875 pint_t p = addr;
4876 pint_t result;
4877
4878 // first get value
4879 switch (encoding & 0x0F) {
4880 case DW_EH_PE_ptr:
4881 result = getP(addr);
4882 p += sizeof(pint_t);
4883 addr = (pint_t)p;
4884 break;
4885 case DW_EH_PE_uleb128:
4886 result = getULEB128(addr, end);
4887 break;
4888 case DW_EH_PE_udata2:
4889 result = get16(addr);
4890 p += 2;
4891 addr = (pint_t)p;
4892 break;
4893 case DW_EH_PE_udata4:
4894 result = get32(addr);
4895 p += 4;
4896 addr = (pint_t)p;
4897 break;
4898 case DW_EH_PE_udata8:
4899 result = get64(addr);
4900 p += 8;
4901 addr = (pint_t)p;
4902 break;
4903 case DW_EH_PE_sleb128:
4904 result = getSLEB128(addr, end);
4905 break;
4906 case DW_EH_PE_sdata2:
4907 result = (int16_t)get16(addr);
4908 p += 2;
4909 addr = (pint_t)p;
4910 break;
4911 case DW_EH_PE_sdata4:
4912 result = (int32_t)get32(addr);
4913 p += 4;
4914 addr = (pint_t)p;
4915 break;
4916 case DW_EH_PE_sdata8:
4917 result = get64(addr);
4918 p += 8;
4919 addr = (pint_t)p;
4920 break;
4921 default:
4922 throwf("ObjectFileAddressSpace<A>::getEncodedP() encoding 0x%08X not supported", encoding);
4923 }
4924
4925 // then add relative offset
4926 switch ( encoding & 0x70 ) {
4927 case DW_EH_PE_absptr:
4928 // do nothing
4929 break;
4930 case DW_EH_PE_pcrel:
4931 result += startAddr;
4932 break;
4933 case DW_EH_PE_textrel:
4934 throw "DW_EH_PE_textrel pointer encoding not supported";
4935 break;
4936 case DW_EH_PE_datarel:
4937 throw "DW_EH_PE_datarel pointer encoding not supported";
4938 break;
4939 case DW_EH_PE_funcrel:
4940 throw "DW_EH_PE_funcrel pointer encoding not supported";
4941 break;
4942 case DW_EH_PE_aligned:
4943 throw "DW_EH_PE_aligned pointer encoding not supported";
4944 break;
4945 default:
4946 throwf("ObjectFileAddressSpace<A>::getEncodedP() encoding 0x%08X not supported", encoding);
4947 break;
4948 }
4949
4950 // Note: DW_EH_PE_indirect is only used in CIEs to refernce the personality pointer
4951 // When parsing .o files that pointer contains zero, so we don't to return that.
4952 // Instead we skip the dereference and return the address of the pointer.
4953 // if ( encoding & DW_EH_PE_indirect )
4954 // result = getP(result);
4955
4956 return result;
4957 }
4958
4959 template <>
4960 const char* CUSection<x86_64>::personalityName(class Parser<x86_64>& parser, const macho_relocation_info<x86_64::P>* reloc)
4961 {
4962 if ( reloc->r_extern() ) {
4963 assert((reloc->r_type() == X86_64_RELOC_UNSIGNED) && "wrong reloc type on personality column in __compact_unwind section");
4964 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
4965 return parser.nameFromSymbol(sym);
4966 }
4967 else {
4968 const pint_t* content = (pint_t*)(this->file().fileContent() + this->_machOSection->offset() + reloc->r_address());
4969 pint_t personalityAddr = *content;
4970 assert((parser.sectionForAddress(personalityAddr)->type() == ld::Section::typeCode) && "personality column in __compact_unwind section is not pointer to function");
4971 // atoms may not be constructed yet, so scan symbol table for labels
4972 const char* name = parser.scanSymbolTableForAddress(personalityAddr);
4973 return name;
4974 }
4975 }
4976
4977 template <>
4978 const char* CUSection<x86>::personalityName(class Parser<x86>& parser, const macho_relocation_info<x86::P>* reloc)
4979 {
4980 if ( reloc->r_extern() ) {
4981 assert((reloc->r_type() == GENERIC_RELOC_VANILLA) && "wrong reloc type on personality column in __compact_unwind section");
4982 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
4983 return parser.nameFromSymbol(sym);
4984 }
4985 else {
4986 // support __LD, __compact_unwind personality entries which are pointer to personality non-lazy pointer
4987 const pint_t* content = (pint_t*)(this->file().fileContent() + this->_machOSection->offset() + reloc->r_address());
4988 pint_t nlPointerAddr = *content;
4989 Section<x86>* nlSection = parser.sectionForAddress(nlPointerAddr);
4990 if ( nlSection->type() == ld::Section::typeCode ) {
4991 // personality function is defined in this .o file, so this is a direct reference to it
4992 // atoms may not be constructed yet, so scan symbol table for labels
4993 const char* name = parser.scanSymbolTableForAddress(nlPointerAddr);
4994 return name;
4995 }
4996 else {
4997 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(nlPointerAddr, nlSection->machoSection());
4998 const macho_nlist<P>& nlSymbol = parser.symbolFromIndex(symIndex);
4999 return parser.nameFromSymbol(nlSymbol);
5000 }
5001 }
5002 }
5003
5004 #if SUPPORT_ARCH_arm64
5005 template <>
5006 const char* CUSection<arm64>::personalityName(class Parser<arm64>& parser, const macho_relocation_info<arm64::P>* reloc)
5007 {
5008 if ( reloc->r_extern() ) {
5009 assert((reloc->r_type() == ARM64_RELOC_UNSIGNED) && "wrong reloc type on personality column in __compact_unwind section");
5010 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
5011 return parser.nameFromSymbol(sym);
5012 }
5013 else {
5014 const pint_t* content = (pint_t*)(this->file().fileContent() + this->_machOSection->offset() + reloc->r_address());
5015 pint_t personalityAddr = *content;
5016 Section<arm64>* personalitySection = parser.sectionForAddress(personalityAddr);
5017 (void)personalitySection;
5018 assert((personalitySection->type() == ld::Section::typeCode) && "personality column in __compact_unwind section is not pointer to function");
5019 // atoms may not be constructed yet, so scan symbol table for labels
5020 const char* name = parser.scanSymbolTableForAddress(personalityAddr);
5021 return name;
5022 }
5023 }
5024 #endif
5025
5026
5027 #if SUPPORT_ARCH_arm_any
5028 template <>
5029 const char* CUSection<arm>::personalityName(class Parser<arm>& parser, const macho_relocation_info<arm::P>* reloc)
5030 {
5031 if ( reloc->r_extern() ) {
5032 assert((reloc->r_type() == ARM_RELOC_VANILLA) && "wrong reloc type on personality column in __compact_unwind section");
5033 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
5034 return parser.nameFromSymbol(sym);
5035 }
5036 else {
5037 // support __LD, __compact_unwind personality entries which are pointer to personality non-lazy pointer
5038 const pint_t* content = (pint_t*)(this->file().fileContent() + this->_machOSection->offset() + reloc->r_address());
5039 pint_t nlPointerAddr = *content;
5040 Section<arm>* nlSection = parser.sectionForAddress(nlPointerAddr);
5041 if ( nlSection->type() == ld::Section::typeCode ) {
5042 // personality function is defined in this .o file, so this is a direct reference to it
5043 // atoms may not be constructed yet, so scan symbol table for labels
5044 const char* name = parser.scanSymbolTableForAddress(nlPointerAddr);
5045 return name;
5046 }
5047 else {
5048 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(nlPointerAddr, nlSection->machoSection());
5049 const macho_nlist<P>& nlSymbol = parser.symbolFromIndex(symIndex);
5050 return parser.nameFromSymbol(nlSymbol);
5051 }
5052 }
5053 }
5054 #endif
5055
5056
5057 template <typename A>
5058 const char* CUSection<A>::personalityName(class Parser<A>& parser, const macho_relocation_info<P>* reloc)
5059 {
5060 return NULL;
5061 }
5062
5063 template <>
5064 bool CUSection<x86>::encodingMeansUseDwarf(compact_unwind_encoding_t enc)
5065 {
5066 return ((enc & UNWIND_X86_MODE_MASK) == UNWIND_X86_MODE_DWARF);
5067 }
5068
5069 template <>
5070 bool CUSection<x86_64>::encodingMeansUseDwarf(compact_unwind_encoding_t enc)
5071 {
5072 return ((enc & UNWIND_X86_64_MODE_MASK) == UNWIND_X86_64_MODE_DWARF);
5073 }
5074
5075 #if SUPPORT_ARCH_arm_any
5076 template <>
5077 bool CUSection<arm>::encodingMeansUseDwarf(compact_unwind_encoding_t enc)
5078 {
5079 return ((enc & UNWIND_ARM_MODE_MASK) == UNWIND_ARM_MODE_DWARF);
5080 }
5081 #endif
5082
5083 #if SUPPORT_ARCH_arm64
5084 template <>
5085 bool CUSection<arm64>::encodingMeansUseDwarf(compact_unwind_encoding_t enc)
5086 {
5087 return ((enc & UNWIND_ARM64_MODE_MASK) == UNWIND_ARM64_MODE_DWARF);
5088 }
5089 #endif
5090
5091
5092 template <typename A>
5093 int CUSection<A>::infoSorter(const void* l, const void* r)
5094 {
5095 // sort references by symbol index, then address
5096 const Info* left = (Info*)l;
5097 const Info* right = (Info*)r;
5098 if ( left->functionSymbolIndex == right->functionSymbolIndex )
5099 return (left->functionStartAddress - right->functionStartAddress);
5100 else
5101 return (left->functionSymbolIndex - right->functionSymbolIndex);
5102 }
5103
5104 template <typename A>
5105 void CUSection<A>::parse(class Parser<A>& parser, uint32_t cnt, Info array[])
5106 {
5107 // walk section content and copy to Info array
5108 const macho_compact_unwind_entry<P>* const entries = (macho_compact_unwind_entry<P>*)(this->file().fileContent() + this->_machOSection->offset());
5109 for (uint32_t i=0; i < cnt; ++i) {
5110 Info* info = &array[i];
5111 const macho_compact_unwind_entry<P>* entry = &entries[i];
5112 info->functionStartAddress = entry->codeStart();
5113 info->functionSymbolIndex = 0xFFFFFFFF;
5114 info->rangeLength = entry->codeLen();
5115 info->compactUnwindInfo = entry->compactUnwindInfo();
5116 info->personality = NULL;
5117 info->lsdaAddress = entry->lsda();
5118 info->function = NULL;
5119 info->lsda = NULL;
5120 if ( (info->compactUnwindInfo & UNWIND_PERSONALITY_MASK) != 0 )
5121 warning("no bits should be set in UNWIND_PERSONALITY_MASK of compact unwind encoding in __LD,__compact_unwind section");
5122 if ( info->lsdaAddress != 0 ) {
5123 info->compactUnwindInfo |= UNWIND_HAS_LSDA;
5124 }
5125 }
5126
5127 // scan relocs, extern relocs are needed for personality references (possibly for function/lsda refs??)
5128 const uint32_t sectionSize = this->_machOSection->size();
5129 const macho_relocation_info<P>* relocs = (macho_relocation_info<P>*)(this->file().fileContent() + this->_machOSection->reloff());
5130 const macho_relocation_info<P>* relocsEnd = &relocs[this->_machOSection->nreloc()];
5131 for (const macho_relocation_info<P>* reloc = relocs; reloc < relocsEnd; ++reloc) {
5132 if ( reloc->r_address() & R_SCATTERED )
5133 continue;
5134 if ( reloc->r_address() > sectionSize )
5135 throwf("malformed __compact_unwind relocation, offset (0x%08X) is beyond end of section,", reloc->r_address());
5136 if ( reloc->r_extern() ) {
5137 // only expect external relocs on some colummns
5138 if ( (reloc->r_address() % sizeof(macho_compact_unwind_entry<P>)) == macho_compact_unwind_entry<P>::personalityFieldOffset() ) {
5139 uint32_t entryIndex = reloc->r_address() / sizeof(macho_compact_unwind_entry<P>);
5140 array[entryIndex].personality = this->personalityName(parser, reloc);
5141 }
5142 else if ( (reloc->r_address() % sizeof(macho_compact_unwind_entry<P>)) == macho_compact_unwind_entry<P>::lsdaFieldOffset() ) {
5143 uint32_t entryIndex = reloc->r_address() / sizeof(macho_compact_unwind_entry<P>);
5144 const macho_nlist<P>& lsdaSym = parser.symbolFromIndex(reloc->r_symbolnum());
5145 if ( (lsdaSym.n_type() & N_TYPE) == N_SECT )
5146 array[entryIndex].lsdaAddress = lsdaSym.n_value();
5147 else
5148 warning("unexpected extern relocation to lsda in __compact_unwind section");
5149 }
5150 else if ( (reloc->r_address() % sizeof(macho_compact_unwind_entry<P>)) == macho_compact_unwind_entry<P>::codeStartFieldOffset() ) {
5151 uint32_t entryIndex = reloc->r_address() / sizeof(macho_compact_unwind_entry<P>);
5152 array[entryIndex].functionSymbolIndex = reloc->r_symbolnum();
5153 array[entryIndex].functionStartAddress += parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
5154 }
5155 else {
5156 warning("unexpected extern relocation in __compact_unwind section");
5157 }
5158 }
5159 else {
5160 if ( (reloc->r_address() % sizeof(macho_compact_unwind_entry<P>)) == macho_compact_unwind_entry<P>::personalityFieldOffset() ) {
5161 uint32_t entryIndex = reloc->r_address() / sizeof(macho_compact_unwind_entry<P>);
5162 array[entryIndex].personality = this->personalityName(parser, reloc);
5163 }
5164 }
5165 }
5166
5167 // sort array by function start address so unwind infos will be contiguous for a given function
5168 ::qsort(array, cnt, sizeof(Info), infoSorter);
5169 }
5170
5171 template <typename A>
5172 uint32_t CUSection<A>::count()
5173 {
5174 const macho_section<P>* machoSect = this->machoSection();
5175 if ( (machoSect->size() % sizeof(macho_compact_unwind_entry<P>)) != 0 )
5176 throw "malformed __LD,__compact_unwind section, bad length";
5177
5178 return machoSect->size() / sizeof(macho_compact_unwind_entry<P>);
5179 }
5180
5181 template <typename A>
5182 void CUSection<A>::makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays& cus)
5183 {
5184 Info* const arrayStart = cus.cuArray;
5185 Info* const arrayEnd = &cus.cuArray[cus.cuCount];
5186 for (Info* info=arrayStart; info < arrayEnd; ++info) {
5187 // find function atom from address
5188 info->function = parser.findAtomByAddress(info->functionStartAddress);
5189 // find lsda atom from address
5190 if ( info->lsdaAddress != 0 ) {
5191 info->lsda = parser.findAtomByAddress(info->lsdaAddress);
5192 // add lsda subordinate
5193 typename Parser<A>::SourceLocation src(info->function, info->functionStartAddress - info->function->objectAddress());
5194 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateLSDA, info->lsda);
5195 }
5196 if ( info->personality != NULL ) {
5197 // add personality subordinate
5198 typename Parser<A>::SourceLocation src(info->function, info->functionStartAddress - info->function->objectAddress());
5199 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinatePersonality, false, info->personality);
5200 }
5201 }
5202
5203 }
5204
5205 template <typename A>
5206 SymboledSection<A>::SymboledSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
5207 : Section<A>(f, s), _type(ld::Atom::typeUnclassified)
5208 {
5209 switch ( s->flags() & SECTION_TYPE ) {
5210 case S_ZEROFILL:
5211 _type = ld::Atom::typeZeroFill;
5212 break;
5213 case S_MOD_INIT_FUNC_POINTERS:
5214 _type = ld::Atom::typeInitializerPointers;
5215 break;
5216 case S_MOD_TERM_FUNC_POINTERS:
5217 _type = ld::Atom::typeTerminatorPointers;
5218 break;
5219 case S_THREAD_LOCAL_VARIABLES:
5220 _type = ld::Atom::typeTLV;
5221 break;
5222 case S_THREAD_LOCAL_ZEROFILL:
5223 _type = ld::Atom::typeTLVZeroFill;
5224 break;
5225 case S_THREAD_LOCAL_REGULAR:
5226 _type = ld::Atom::typeTLVInitialValue;
5227 break;
5228 case S_THREAD_LOCAL_INIT_FUNCTION_POINTERS:
5229 _type = ld::Atom::typeTLVInitializerPointers;
5230 break;
5231 case S_REGULAR:
5232 if ( strncmp(s->sectname(), "__gcc_except_tab", 16) == 0 )
5233 _type = ld::Atom::typeLSDA;
5234 else if ( this->type() == ld::Section::typeInitializerPointers )
5235 _type = ld::Atom::typeInitializerPointers;
5236 break;
5237 }
5238 }
5239
5240
5241 template <typename A>
5242 bool SymboledSection<A>::dontDeadStrip()
5243 {
5244 switch ( _type ) {
5245 case ld::Atom::typeInitializerPointers:
5246 case ld::Atom::typeTerminatorPointers:
5247 return true;
5248 default:
5249 // model an object file without MH_SUBSECTIONS_VIA_SYMBOLS as one in which nothing can be dead stripped
5250 if ( ! this->_file.canScatterAtoms() )
5251 return true;
5252 // call inherited
5253 return Section<A>::dontDeadStrip();
5254 }
5255 return false;
5256 }
5257
5258
5259 template <typename A>
5260 uint32_t SymboledSection<A>::computeAtomCount(class Parser<A>& parser,
5261 struct Parser<A>::LabelAndCFIBreakIterator& it,
5262 const struct Parser<A>::CFI_CU_InfoArrays&)
5263 {
5264 const pint_t startAddr = this->_machOSection->addr();
5265 const pint_t endAddr = startAddr + this->_machOSection->size();
5266 const uint32_t sectNum = this->sectionNum(parser);
5267
5268 uint32_t count = 0;
5269 pint_t addr;
5270 pint_t size;
5271 const macho_nlist<P>* sym;
5272 while ( it.next(parser, *this, sectNum, startAddr, endAddr, &addr, &size, &sym) ) {
5273 ++count;
5274 }
5275 //fprintf(stderr, "computeAtomCount(%s,%s) => %d\n", this->segmentName(), this->sectionName(), count);
5276 return count;
5277 }
5278
5279 template <typename A>
5280 uint32_t SymboledSection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
5281 struct Parser<A>::LabelAndCFIBreakIterator& it,
5282 const struct Parser<A>::CFI_CU_InfoArrays&)
5283 {
5284 this->_beginAtoms = (Atom<A>*)p;
5285
5286 //fprintf(stderr, "SymboledSection::appendAtoms() in section %s\n", this->_machOSection->sectname());
5287 const pint_t startAddr = this->_machOSection->addr();
5288 const pint_t endAddr = startAddr + this->_machOSection->size();
5289 const uint32_t sectNum = this->sectionNum(parser);
5290
5291 uint32_t count = 0;
5292 pint_t addr;
5293 pint_t size;
5294 const macho_nlist<P>* label;
5295 while ( it.next(parser, *this, sectNum, startAddr, endAddr, &addr, &size, &label) ) {
5296 Atom<A>* allocatedSpace = (Atom<A>*)p;
5297 // is break because of label or CFI?
5298 if ( label != NULL ) {
5299 // The size is computed based on the address of the next label (or the end of the section for the last label)
5300 // If there are two labels at the same address, we want them one to be an alias of the other.
5301 // If the label is at the end of a section, it is has zero size, but is not an alias
5302 const bool isAlias = ( (size == 0) && (addr < endAddr) );
5303 new (allocatedSpace) Atom<A>(*this, parser, *label, size, isAlias);
5304 if ( isAlias )
5305 this->_hasAliases = true;
5306 if ( parser.altEntryFromSymbol(*label) )
5307 this->_altEntries.insert(allocatedSpace);
5308 }
5309 else {
5310 ld::Atom::SymbolTableInclusion inclusion = ld::Atom::symbolTableNotIn;
5311 ld::Atom::ContentType ctype = this->contentType();
5312 if ( ctype == ld::Atom::typeLSDA )
5313 inclusion = ld::Atom::symbolTableInWithRandomAutoStripLabel;
5314 new (allocatedSpace) Atom<A>(*this, "anon", addr, size, ld::Atom::definitionRegular, ld::Atom::combineNever,
5315 ld::Atom::scopeTranslationUnit, ctype, inclusion,
5316 this->dontDeadStrip(), false, false, this->alignmentForAddress(addr));
5317 }
5318 p += sizeof(Atom<A>);
5319 ++count;
5320 }
5321
5322 this->_endAtoms = (Atom<A>*)p;
5323 return count;
5324 }
5325
5326
5327 template <>
5328 ld::Atom::SymbolTableInclusion ImplicitSizeSection<arm64>::symbolTableInclusion()
5329 {
5330 return ld::Atom::symbolTableInWithRandomAutoStripLabel;
5331 }
5332
5333
5334 template <typename A>
5335 ld::Atom::SymbolTableInclusion ImplicitSizeSection<A>::symbolTableInclusion()
5336 {
5337 return ld::Atom::symbolTableNotIn;
5338 }
5339
5340
5341 template <typename A>
5342 uint32_t ImplicitSizeSection<A>::computeAtomCount(class Parser<A>& parser,
5343 struct Parser<A>::LabelAndCFIBreakIterator& it,
5344 const struct Parser<A>::CFI_CU_InfoArrays&)
5345 {
5346 uint32_t count = 0;
5347 const macho_section<P>* sect = this->machoSection();
5348 const pint_t startAddr = sect->addr();
5349 const pint_t endAddr = startAddr + sect->size();
5350 for (pint_t addr = startAddr; addr < endAddr; addr += elementSizeAtAddress(addr) ) {
5351 if ( useElementAt(parser, it, addr) )
5352 ++count;
5353 }
5354 if ( it.fileHasOverlappingSymbols && (sect->size() != 0) && (this->combine(parser, startAddr) == ld::Atom::combineByNameAndContent) ) {
5355 // if there are multiple labels in this section for the same address, then clone them into multi atoms
5356 pint_t prevSymbolAddr = (pint_t)(-1);
5357 uint8_t prevSymbolSectNum = 0;
5358 bool prevIgnore = false;
5359 for(uint32_t i=0; i < it.sortedSymbolCount; ++i) {
5360 const macho_nlist<P>& sym = parser.symbolFromIndex(it.sortedSymbolIndexes[i]);
5361 const pint_t symbolAddr = sym.n_value();
5362 const uint8_t symbolSectNum = sym.n_sect();
5363 const bool ignore = this->ignoreLabel(parser.nameFromSymbol(sym));
5364 if ( !ignore && !prevIgnore && (symbolAddr == prevSymbolAddr) && (prevSymbolSectNum == symbolSectNum) && (symbolSectNum == this->sectionNum(parser)) ) {
5365 ++count;
5366 }
5367 prevSymbolAddr = symbolAddr;
5368 prevSymbolSectNum = symbolSectNum;
5369 prevIgnore = ignore;
5370 }
5371 }
5372 return count;
5373 }
5374
5375 template <typename A>
5376 uint32_t ImplicitSizeSection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
5377 struct Parser<A>::LabelAndCFIBreakIterator& it,
5378 const struct Parser<A>::CFI_CU_InfoArrays&)
5379 {
5380 this->_beginAtoms = (Atom<A>*)p;
5381
5382 const macho_section<P>* sect = this->machoSection();
5383 const pint_t startAddr = sect->addr();
5384 const pint_t endAddr = startAddr + sect->size();
5385 const uint32_t sectNum = this->sectionNum(parser);
5386 //fprintf(stderr, "ImplicitSizeSection::appendAtoms() in section %s\n", sect->sectname());
5387 uint32_t count = 0;
5388 pint_t foundAddr;
5389 pint_t size;
5390 const macho_nlist<P>* foundLabel;
5391 Atom<A>* allocatedSpace;
5392 while ( it.next(parser, *this, sectNum, startAddr, endAddr, &foundAddr, &size, &foundLabel) ) {
5393 if ( foundLabel != NULL ) {
5394 bool skip = false;
5395 pint_t labeledAtomSize = this->elementSizeAtAddress(foundAddr);
5396 allocatedSpace = (Atom<A>*)p;
5397 if ( this->ignoreLabel(parser.nameFromSymbol(*foundLabel)) ) {
5398 if ( size == 0 ) {
5399 // <rdar://problem/10018737>
5400 // a size of zero means there is another label at same location
5401 // and we are supposed to ignore this label
5402 skip = true;
5403 }
5404 else {
5405 //fprintf(stderr, " 0x%08llX make annon, size=%lld\n", (uint64_t)foundAddr, (uint64_t)size);
5406 new (allocatedSpace) Atom<A>(*this, this->unlabeledAtomName(parser, foundAddr), foundAddr,
5407 this->elementSizeAtAddress(foundAddr), this->definition(),
5408 this->combine(parser, foundAddr), this->scopeAtAddress(parser, foundAddr),
5409 this->contentType(), this->symbolTableInclusion(),
5410 this->dontDeadStrip(), false, false, this->alignmentForAddress(foundAddr));
5411 }
5412 }
5413 else {
5414 // make named atom for label
5415 //fprintf(stderr, " 0x%08llX make labeled\n", (uint64_t)foundAddr);
5416 new (allocatedSpace) Atom<A>(*this, parser, *foundLabel, labeledAtomSize);
5417 }
5418 if ( !skip ) {
5419 ++count;
5420 p += sizeof(Atom<A>);
5421 foundAddr += labeledAtomSize;
5422 size -= labeledAtomSize;
5423 }
5424 }
5425 // some number of anonymous atoms
5426 for (pint_t addr = foundAddr; addr < (foundAddr+size); addr += elementSizeAtAddress(addr) ) {
5427 // make anon atoms for area before label
5428 if ( this->useElementAt(parser, it, addr) ) {
5429 //fprintf(stderr, " 0x%08llX make annon, size=%lld\n", (uint64_t)addr, (uint64_t)elementSizeAtAddress(addr));
5430 allocatedSpace = (Atom<A>*)p;
5431 new (allocatedSpace) Atom<A>(*this, this->unlabeledAtomName(parser, addr), addr, this->elementSizeAtAddress(addr),
5432 this->definition(), this->combine(parser, addr), this->scopeAtAddress(parser, addr),
5433 this->contentType(), this->symbolTableInclusion(),
5434 this->dontDeadStrip(), false, false, this->alignmentForAddress(addr));
5435 ++count;
5436 p += sizeof(Atom<A>);
5437 }
5438 }
5439 }
5440
5441 this->_endAtoms = (Atom<A>*)p;
5442
5443 return count;
5444 }
5445
5446 template <typename A>
5447 bool Literal4Section<A>::ignoreLabel(const char* label) const
5448 {
5449 return (label[0] == 'L') || (label[0] == 'l');
5450 }
5451
5452 template <typename A>
5453 unsigned long Literal4Section<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5454 {
5455 const uint32_t* literalContent = (uint32_t*)atom->contentPointer();
5456 return *literalContent;
5457 }
5458
5459 template <typename A>
5460 bool Literal4Section<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5461 const ld::IndirectBindingTable& ind) const
5462 {
5463 assert(this->type() == rhs.section().type());
5464 const uint32_t* literalContent = (uint32_t*)atom->contentPointer();
5465
5466 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5467 assert(rhsAtom != NULL);
5468 if ( rhsAtom != NULL ) {
5469 const uint32_t* rhsLiteralContent = (uint32_t*)rhsAtom->contentPointer();
5470 return (*literalContent == *rhsLiteralContent);
5471 }
5472 return false;
5473 }
5474
5475
5476 template <typename A>
5477 bool Literal8Section<A>::ignoreLabel(const char* label) const
5478 {
5479 return (label[0] == 'L') || (label[0] == 'l');
5480 }
5481
5482 template <typename A>
5483 unsigned long Literal8Section<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5484 {
5485 #if __LP64__
5486 const uint64_t* literalContent = (uint64_t*)atom->contentPointer();
5487 return *literalContent;
5488 #else
5489 unsigned long hash = 5381;
5490 const uint8_t* byteContent = atom->contentPointer();
5491 for (int i=0; i < 8; ++i) {
5492 hash = hash * 33 + byteContent[i];
5493 }
5494 return hash;
5495 #endif
5496 }
5497
5498 template <typename A>
5499 bool Literal8Section<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5500 const ld::IndirectBindingTable& ind) const
5501 {
5502 if ( rhs.section().type() != ld::Section::typeLiteral8 )
5503 return false;
5504 assert(this->type() == rhs.section().type());
5505 const uint64_t* literalContent = (uint64_t*)atom->contentPointer();
5506
5507 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5508 assert(rhsAtom != NULL);
5509 if ( rhsAtom != NULL ) {
5510 const uint64_t* rhsLiteralContent = (uint64_t*)rhsAtom->contentPointer();
5511 return (*literalContent == *rhsLiteralContent);
5512 }
5513 return false;
5514 }
5515
5516 template <typename A>
5517 bool Literal16Section<A>::ignoreLabel(const char* label) const
5518 {
5519 return (label[0] == 'L') || (label[0] == 'l');
5520 }
5521
5522 template <typename A>
5523 unsigned long Literal16Section<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5524 {
5525 unsigned long hash = 5381;
5526 const uint8_t* byteContent = atom->contentPointer();
5527 for (int i=0; i < 16; ++i) {
5528 hash = hash * 33 + byteContent[i];
5529 }
5530 return hash;
5531 }
5532
5533 template <typename A>
5534 bool Literal16Section<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5535 const ld::IndirectBindingTable& ind) const
5536 {
5537 if ( rhs.section().type() != ld::Section::typeLiteral16 )
5538 return false;
5539 assert(this->type() == rhs.section().type());
5540 const uint64_t* literalContent = (uint64_t*)atom->contentPointer();
5541
5542 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5543 assert(rhsAtom != NULL);
5544 if ( rhsAtom != NULL ) {
5545 const uint64_t* rhsLiteralContent = (uint64_t*)rhsAtom->contentPointer();
5546 return ((literalContent[0] == rhsLiteralContent[0]) && (literalContent[1] == rhsLiteralContent[1]));
5547 }
5548 return false;
5549 }
5550
5551
5552
5553 template <typename A>
5554 typename A::P::uint_t CStringSection<A>::elementSizeAtAddress(pint_t addr)
5555 {
5556 const macho_section<P>* sect = this->machoSection();
5557 const char* stringContent = (char*)(this->file().fileContent() + sect->offset() + addr - sect->addr());
5558 return strlen(stringContent) + 1;
5559 }
5560
5561 template <typename A>
5562 bool CStringSection<A>::useElementAt(Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it, pint_t addr)
5563 {
5564 return true;
5565 }
5566
5567 template <typename A>
5568 bool CStringSection<A>::ignoreLabel(const char* label) const
5569 {
5570 return (label[0] == 'L') || (label[0] == 'l');
5571 }
5572
5573
5574 template <typename A>
5575 Atom<A>* CStringSection<A>::findAtomByAddress(pint_t addr)
5576 {
5577 Atom<A>* result = this->findContentAtomByAddress(addr, this->_beginAtoms, this->_endAtoms);
5578 return result;
5579 }
5580
5581 template <typename A>
5582 unsigned long CStringSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5583 {
5584 unsigned long hash = 5381;
5585 const char* stringContent = (char*)atom->contentPointer();
5586 for (const char* s = stringContent; *s != '\0'; ++s) {
5587 hash = hash * 33 + *s;
5588 }
5589 return hash;
5590 }
5591
5592
5593 template <typename A>
5594 bool CStringSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5595 const ld::IndirectBindingTable& ind) const
5596 {
5597 if ( rhs.section().type() != ld::Section::typeCString )
5598 return false;
5599 assert(this->type() == rhs.section().type());
5600 assert(strcmp(this->sectionName(), rhs.section().sectionName())== 0);
5601 assert(strcmp(this->segmentName(), rhs.section().segmentName())== 0);
5602 const char* stringContent = (char*)atom->contentPointer();
5603
5604 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5605 assert(rhsAtom != NULL);
5606 if ( rhsAtom != NULL ) {
5607 if ( atom->_size != rhsAtom->_size )
5608 return false;
5609 const char* rhsStringContent = (char*)rhsAtom->contentPointer();
5610 return (strcmp(stringContent, rhsStringContent) == 0);
5611 }
5612 return false;
5613 }
5614
5615
5616 template <>
5617 ld::Fixup::Kind NonLazyPointerSection<x86>::fixupKind()
5618 {
5619 return ld::Fixup::kindStoreLittleEndian32;
5620 }
5621
5622 template <>
5623 ld::Fixup::Kind NonLazyPointerSection<arm>::fixupKind()
5624 {
5625 return ld::Fixup::kindStoreLittleEndian32;
5626 }
5627
5628 template <>
5629 ld::Fixup::Kind NonLazyPointerSection<arm64>::fixupKind()
5630 {
5631 return ld::Fixup::kindStoreLittleEndian64;
5632 }
5633
5634
5635 template <>
5636 void NonLazyPointerSection<x86_64>::makeFixups(class Parser<x86_64>& parser, const struct Parser<x86_64>::CFI_CU_InfoArrays&)
5637 {
5638 assert(0 && "x86_64 should not have non-lazy-pointer sections in .o files");
5639 }
5640
5641 template <typename A>
5642 void NonLazyPointerSection<A>::makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&)
5643 {
5644 // add references for each NLP atom based on indirect symbol table
5645 const macho_section<P>* sect = this->machoSection();
5646 const pint_t endAddr = sect->addr() + sect->size();
5647 for( pint_t addr = sect->addr(); addr < endAddr; addr += sizeof(pint_t)) {
5648 typename Parser<A>::SourceLocation src;
5649 typename Parser<A>::TargetDesc target;
5650 src.atom = this->findAtomByAddress(addr);
5651 src.offsetInAtom = 0;
5652 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(addr, sect);
5653 target.atom = NULL;
5654 target.name = NULL;
5655 target.weakImport = false;
5656 target.addend = 0;
5657 if ( symIndex == INDIRECT_SYMBOL_LOCAL ) {
5658 // use direct reference for local symbols
5659 const pint_t* nlpContent = (pint_t*)(this->file().fileContent() + sect->offset() + addr - sect->addr());
5660 pint_t targetAddr = P::getP(*nlpContent);
5661 target.atom = parser.findAtomByAddress(targetAddr);
5662 target.weakImport = false;
5663 target.addend = (targetAddr - target.atom->objectAddress());
5664 // <rdar://problem/8385011> if pointer to thumb function, mask of thumb bit (not an addend of +1)
5665 if ( target.atom->isThumb() )
5666 target.addend &= (-2);
5667 assert(src.atom->combine() == ld::Atom::combineNever);
5668 }
5669 else {
5670 const macho_nlist<P>& sym = parser.symbolFromIndex(symIndex);
5671 // use direct reference for local symbols
5672 if ( ((sym.n_type() & N_TYPE) == N_SECT) && ((sym.n_type() & N_EXT) == 0) ) {
5673 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
5674 assert(src.atom->combine() == ld::Atom::combineNever);
5675 }
5676 else {
5677 target.name = parser.nameFromSymbol(sym);
5678 target.weakImport = parser.weakImportFromSymbol(sym);
5679 assert(src.atom->combine() == ld::Atom::combineByNameAndReferences);
5680 }
5681 }
5682 parser.addFixups(src, this->fixupKind(), target);
5683 }
5684 }
5685
5686 template <typename A>
5687 ld::Atom::Combine NonLazyPointerSection<A>::combine(Parser<A>& parser, pint_t addr)
5688 {
5689 const macho_section<P>* sect = this->machoSection();
5690 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(addr, sect);
5691 if ( symIndex == INDIRECT_SYMBOL_LOCAL)
5692 return ld::Atom::combineNever;
5693
5694 // don't coalesce non-lazy-pointers to local symbols
5695 const macho_nlist<P>& sym = parser.symbolFromIndex(symIndex);
5696 if ( ((sym.n_type() & N_TYPE) == N_SECT) && ((sym.n_type() & N_EXT) == 0) )
5697 return ld::Atom::combineNever;
5698
5699 return ld::Atom::combineByNameAndReferences;
5700 }
5701
5702 template <typename A>
5703 const char* NonLazyPointerSection<A>::targetName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind)
5704 {
5705 assert(atom->combine() == ld::Atom::combineByNameAndReferences);
5706 assert(atom->fixupCount() == 1);
5707 ld::Fixup::iterator fit = atom->fixupsBegin();
5708 const char* name = NULL;
5709 switch ( fit->binding ) {
5710 case ld::Fixup::bindingByNameUnbound:
5711 name = fit->u.name;
5712 break;
5713 case ld::Fixup::bindingByContentBound:
5714 name = fit->u.target->name();
5715 break;
5716 case ld::Fixup::bindingsIndirectlyBound:
5717 name = ind.indirectName(fit->u.bindingIndex);
5718 break;
5719 default:
5720 assert(0);
5721 }
5722 assert(name != NULL);
5723 return name;
5724 }
5725
5726 template <typename A>
5727 unsigned long NonLazyPointerSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5728 {
5729 assert(atom->combine() == ld::Atom::combineByNameAndReferences);
5730 unsigned long hash = 9508;
5731 for (const char* s = this->targetName(atom, ind); *s != '\0'; ++s) {
5732 hash = hash * 33 + *s;
5733 }
5734 return hash;
5735 }
5736
5737 template <typename A>
5738 bool NonLazyPointerSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5739 const ld::IndirectBindingTable& indirectBindingTable) const
5740 {
5741 if ( rhs.section().type() != ld::Section::typeNonLazyPointer )
5742 return false;
5743 assert(this->type() == rhs.section().type());
5744 // there can be many non-lazy pointer in different section names
5745 // we only want to coalesce in same section name
5746 if ( *this != rhs.section() )
5747 return false;
5748 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5749 assert(rhsAtom != NULL);
5750 const char* thisName = this->targetName(atom, indirectBindingTable);
5751 const char* rhsName = this->targetName(rhsAtom, indirectBindingTable);
5752 return (strcmp(thisName, rhsName) == 0);
5753 }
5754
5755 template <typename A>
5756 ld::Atom::Scope NonLazyPointerSection<A>::scopeAtAddress(Parser<A>& parser, pint_t addr)
5757 {
5758 const macho_section<P>* sect = this->machoSection();
5759 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(addr, sect);
5760 if ( symIndex == INDIRECT_SYMBOL_LOCAL)
5761 return ld::Atom::scopeTranslationUnit;
5762 else
5763 return ld::Atom::scopeLinkageUnit;
5764 }
5765
5766
5767
5768 template <typename A>
5769 ld::Atom::Combine TLVPointerSection<A>::combine(Parser<A>& parser, pint_t addr)
5770 {
5771 return ld::Atom::combineByNameAndReferences;
5772 }
5773
5774 template <>
5775 void TLVPointerSection<arm>::makeFixups(class Parser<arm>& parser, const struct Parser<arm>::CFI_CU_InfoArrays&)
5776 {
5777 // add references for each thread local pointer atom based on indirect symbol table
5778 const macho_section<P>* sect = this->machoSection();
5779 const pint_t endAddr = sect->addr() + sect->size();
5780 for (pint_t addr = sect->addr(); addr < endAddr; addr += sizeof(pint_t)) {
5781 typename Parser<arm>::SourceLocation src;
5782 typename Parser<arm>::TargetDesc target;
5783 src.atom = this->findAtomByAddress(addr);
5784 src.offsetInAtom = 0;
5785 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(addr, sect);
5786 target.atom = NULL;
5787 target.name = NULL;
5788 target.weakImport = false;
5789 target.addend = 0;
5790 if ( symIndex == INDIRECT_SYMBOL_LOCAL ) {
5791 throwf("unexpected INDIRECT_SYMBOL_LOCAL in section %s", this->sectionName());
5792 }
5793 else {
5794 const macho_nlist<P>& sym = parser.symbolFromIndex(symIndex);
5795 // use direct reference for local symbols
5796 if ( ((sym.n_type() & N_TYPE) == N_SECT) && ((sym.n_type() & N_EXT) == 0) ) {
5797 throwf("unexpected pointer to local symbol in section %s", this->sectionName());
5798 }
5799 else {
5800 target.name = parser.nameFromSymbol(sym);
5801 target.weakImport = parser.weakImportFromSymbol(sym);
5802 assert(src.atom->combine() == ld::Atom::combineByNameAndReferences);
5803 }
5804 }
5805 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
5806 }
5807 }
5808
5809 template <typename A>
5810 void TLVPointerSection<A>::makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&)
5811 {
5812 assert(0 && "should not have thread-local-pointer sections in .o files");
5813 }
5814
5815
5816 template <typename A>
5817 const char* TLVPointerSection<A>::targetName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind, bool* isStatic)
5818 {
5819 assert(atom->combine() == ld::Atom::combineByNameAndReferences);
5820 assert(atom->fixupCount() == 1);
5821 *isStatic = false;
5822 ld::Fixup::iterator fit = atom->fixupsBegin();
5823 const char* name = NULL;
5824 switch ( fit->binding ) {
5825 case ld::Fixup::bindingByNameUnbound:
5826 name = fit->u.name;
5827 break;
5828 case ld::Fixup::bindingByContentBound:
5829 name = fit->u.target->name();
5830 break;
5831 case ld::Fixup::bindingsIndirectlyBound:
5832 name = ind.indirectName(fit->u.bindingIndex);
5833 break;
5834 case ld::Fixup::bindingDirectlyBound:
5835 name = fit->u.target->name();
5836 *isStatic = (fit->u.target->scope() == ld::Atom::scopeTranslationUnit);
5837 break;
5838 default:
5839 assert(0);
5840 }
5841 assert(name != NULL);
5842 return name;
5843 }
5844
5845 template <typename A>
5846 unsigned long TLVPointerSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5847 {
5848 assert(atom->combine() == ld::Atom::combineByNameAndReferences);
5849 unsigned long hash = 9508;
5850 bool isStatic;
5851 for (const char* s = this->targetName(atom, ind, &isStatic); *s != '\0'; ++s) {
5852 hash = hash * 33 + *s;
5853 }
5854 return hash;
5855 }
5856
5857 template <typename A>
5858 bool TLVPointerSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5859 const ld::IndirectBindingTable& indirectBindingTable) const
5860 {
5861 if ( rhs.section().type() != ld::Section::typeTLVPointers )
5862 return false;
5863 assert(this->type() == rhs.section().type());
5864 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5865 assert(rhsAtom != NULL);
5866 bool thisIsStatic;
5867 bool rhsIsStatic;
5868 const char* thisName = this->targetName(atom, indirectBindingTable, &thisIsStatic);
5869 const char* rhsName = this->targetName(rhsAtom, indirectBindingTable, &rhsIsStatic);
5870 return !thisIsStatic && !rhsIsStatic && (strcmp(thisName, rhsName) == 0);
5871 }
5872
5873
5874 template <typename A>
5875 const uint8_t* CFStringSection<A>::targetContent(const class Atom<A>* atom, const ld::IndirectBindingTable& ind,
5876 ContentType* ct, unsigned int* count)
5877 {
5878 *ct = contentUnknown;
5879 for (ld::Fixup::iterator fit=atom->fixupsBegin(), end=atom->fixupsEnd(); fit != end; ++fit) {
5880 const ld::Atom* targetAtom = NULL;
5881 switch ( fit->binding ) {
5882 case ld::Fixup::bindingByNameUnbound:
5883 // ignore reference to ___CFConstantStringClassReference
5884 // we are just looking for reference to backing string data
5885 assert(fit->offsetInAtom == 0);
5886 assert(strcmp(fit->u.name, "___CFConstantStringClassReference") == 0);
5887 break;
5888 case ld::Fixup::bindingDirectlyBound:
5889 case ld::Fixup::bindingByContentBound:
5890 targetAtom = fit->u.target;
5891 break;
5892 case ld::Fixup::bindingsIndirectlyBound:
5893 targetAtom = ind.indirectAtom(fit->u.bindingIndex);
5894 break;
5895 default:
5896 assert(0 && "bad binding type");
5897 }
5898 assert(targetAtom != NULL);
5899 const Atom<A>* target = dynamic_cast<const Atom<A>*>(targetAtom);
5900 if ( targetAtom->section().type() == ld::Section::typeCString ) {
5901 *ct = contentUTF8;
5902 *count = targetAtom->size();
5903 }
5904 else if ( targetAtom->section().type() == ld::Section::typeUTF16Strings ) {
5905 *ct = contentUTF16;
5906 *count = (targetAtom->size()+1)/2; // round up incase of buggy compiler that has only one trailing zero byte
5907 }
5908 else {
5909 *ct = contentUnknown;
5910 *count = 0;
5911 return NULL;
5912 }
5913 return target->contentPointer();
5914 }
5915 assert(0);
5916 return NULL;
5917 }
5918
5919 template <typename A>
5920 unsigned long CFStringSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5921 {
5922 // base hash of CFString on hash of cstring it wraps
5923 ContentType cType;
5924 unsigned long hash;
5925 unsigned int charCount;
5926 const uint8_t* content = this->targetContent(atom, ind, &cType, &charCount);
5927 switch ( cType ) {
5928 case contentUTF8:
5929 hash = 9408;
5930 for (const char* s = (char*)content; *s != '\0'; ++s) {
5931 hash = hash * 33 + *s;
5932 }
5933 return hash;
5934 case contentUTF16:
5935 hash = 407955;
5936 --charCount; // don't add last 0x0000 to hash because some buggy compilers only have trailing single byte
5937 for (const uint16_t* s = (uint16_t*)content; charCount > 0; ++s, --charCount) {
5938 hash = hash * 1025 + *s;
5939 }
5940 return hash;
5941 case contentUnknown:
5942 // <rdar://problem/14134211> For malformed CFStrings, hash to address of atom so they have unique hashes
5943 return ULONG_MAX - (unsigned long)(atom);
5944 }
5945 return 0;
5946 }
5947
5948
5949 template <typename A>
5950 bool CFStringSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5951 const ld::IndirectBindingTable& indirectBindingTable) const
5952 {
5953 if ( atom == &rhs )
5954 return true;
5955 if ( rhs.section().type() != ld::Section::typeCFString)
5956 return false;
5957 assert(this->type() == rhs.section().type());
5958 assert(strcmp(this->sectionName(), "__cfstring") == 0);
5959
5960 ContentType thisType;
5961 unsigned int charCount;
5962 const uint8_t* cstringContent = this->targetContent(atom, indirectBindingTable, &thisType, &charCount);
5963 ContentType rhsType;
5964 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5965 assert(rhsAtom != NULL);
5966 unsigned int rhsCharCount;
5967 const uint8_t* rhsStringContent = this->targetContent(rhsAtom, indirectBindingTable, &rhsType, &rhsCharCount);
5968
5969 if ( thisType != rhsType )
5970 return false;
5971
5972 if ( thisType == contentUnknown )
5973 return false;
5974
5975 if ( rhsType == contentUnknown )
5976 return false;
5977
5978 // no need to compare content of pointers are already the same
5979 if ( cstringContent == rhsStringContent )
5980 return true;
5981
5982 // no need to compare content if size is different
5983 if ( charCount != rhsCharCount )
5984 return false;
5985
5986 switch ( thisType ) {
5987 case contentUTF8:
5988 return (strcmp((char*)cstringContent, (char*)rhsStringContent) == 0);
5989 case contentUTF16:
5990 {
5991 const uint16_t* cstringContent16 = (uint16_t*)cstringContent;
5992 const uint16_t* rhsStringContent16 = (uint16_t*)rhsStringContent;
5993 for (unsigned int i = 0; i < charCount; ++i) {
5994 if ( cstringContent16[i] != rhsStringContent16[i] )
5995 return false;
5996 }
5997 return true;
5998 }
5999 case contentUnknown:
6000 return false;
6001 }
6002 return false;
6003 }
6004
6005
6006 template <typename A>
6007 typename A::P::uint_t ObjC1ClassSection<A>::elementSizeAtAddress(pint_t addr)
6008 {
6009 // nominal size for each class is 48 bytes, but sometimes the compiler
6010 // over aligns and there is padding after class data
6011 const macho_section<P>* sct = this->machoSection();
6012 uint32_t align = 1 << sct->align();
6013 uint32_t size = ((12 * sizeof(pint_t)) + align-1) & (-align);
6014 return size;
6015 }
6016
6017 template <typename A>
6018 const char* ObjC1ClassSection<A>::unlabeledAtomName(Parser<A>& parser, pint_t addr)
6019 {
6020 // 8-bytes into class object is pointer to class name
6021 const macho_section<P>* sct = this->machoSection();
6022 uint32_t classObjcFileOffset = sct->offset() - sct->addr() + addr;
6023 const uint8_t* mappedFileContent = this->file().fileContent();
6024 pint_t nameAddr = P::getP(*((pint_t*)(mappedFileContent+classObjcFileOffset+2*sizeof(pint_t))));
6025
6026 // find section containing string address to get string bytes
6027 const macho_section<P>* const sections = parser.firstMachOSection();
6028 const uint32_t sectionCount = parser.machOSectionCount();
6029 for (uint32_t i=0; i < sectionCount; ++i) {
6030 const macho_section<P>* aSect = &sections[i];
6031 if ( (aSect->addr() <= nameAddr) && (nameAddr < (aSect->addr()+aSect->size())) ) {
6032 assert((aSect->flags() & SECTION_TYPE) == S_CSTRING_LITERALS);
6033 uint32_t nameFileOffset = aSect->offset() - aSect->addr() + nameAddr;
6034 const char* name = (char*)mappedFileContent + nameFileOffset;
6035 // spin through symbol table to find absolute symbol corresponding to this class
6036 for (uint32_t s=0; s < parser.symbolCount(); ++s) {
6037 const macho_nlist<P>& sym = parser.symbolFromIndex(s);
6038 if ( (sym.n_type() & N_TYPE) != N_ABS )
6039 continue;
6040 const char* absName = parser.nameFromSymbol(sym);
6041 if ( strncmp(absName, ".objc_class_name_", 17) == 0 ) {
6042 if ( strcmp(&absName[17], name) == 0 )
6043 return absName;
6044 }
6045 }
6046 assert(0 && "obj class name not found in symbol table");
6047 }
6048 }
6049 assert(0 && "obj class name not found");
6050 return "unknown objc class";
6051 }
6052
6053
6054 template <typename A>
6055 const char* ObjC2ClassRefsSection<A>::targetClassName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
6056 {
6057 assert(atom->fixupCount() == 1);
6058 ld::Fixup::iterator fit = atom->fixupsBegin();
6059 const char* className = NULL;
6060 switch ( fit->binding ) {
6061 case ld::Fixup::bindingByNameUnbound:
6062 className = fit->u.name;
6063 break;
6064 case ld::Fixup::bindingDirectlyBound:
6065 case ld::Fixup::bindingByContentBound:
6066 className = fit->u.target->name();
6067 break;
6068 case ld::Fixup::bindingsIndirectlyBound:
6069 className = ind.indirectName(fit->u.bindingIndex);
6070 break;
6071 default:
6072 assert(0 && "unsupported binding in objc2 class ref section");
6073 }
6074 assert(className != NULL);
6075 return className;
6076 }
6077
6078
6079 template <typename A>
6080 unsigned long ObjC2ClassRefsSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
6081 {
6082 unsigned long hash = 978;
6083 for (const char* s = targetClassName(atom, ind); *s != '\0'; ++s) {
6084 hash = hash * 33 + *s;
6085 }
6086 return hash;
6087 }
6088
6089 template <typename A>
6090 bool ObjC2ClassRefsSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
6091 const ld::IndirectBindingTable& indirectBindingTable) const
6092 {
6093 assert(this->type() == rhs.section().type());
6094 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
6095 assert(rhsAtom != NULL);
6096 const char* thisClassName = targetClassName(atom, indirectBindingTable);
6097 const char* rhsClassName = targetClassName(rhsAtom, indirectBindingTable);
6098 return (strcmp(thisClassName, rhsClassName) == 0);
6099 }
6100
6101
6102 template <typename A>
6103 const char* Objc1ClassReferences<A>::targetCString(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
6104 {
6105 assert(atom->fixupCount() == 2);
6106 ld::Fixup::iterator fit = atom->fixupsBegin();
6107 if ( fit->kind == ld::Fixup::kindSetTargetAddress )
6108 ++fit;
6109 const ld::Atom* targetAtom = NULL;
6110 switch ( fit->binding ) {
6111 case ld::Fixup::bindingByContentBound:
6112 targetAtom = fit->u.target;
6113 break;
6114 case ld::Fixup::bindingsIndirectlyBound:
6115 targetAtom = ind.indirectAtom(fit->u.bindingIndex);
6116 if ( targetAtom == NULL ) {
6117 fprintf(stderr, "missing target named %s\n", ind.indirectName(fit->u.bindingIndex));
6118 }
6119 break;
6120 default:
6121 assert(0);
6122 }
6123 assert(targetAtom != NULL);
6124 const Atom<A>* target = dynamic_cast<const Atom<A>*>(targetAtom);
6125 assert(target != NULL);
6126 return (char*)target->contentPointer();
6127 }
6128
6129
6130 template <typename A>
6131 const char* PointerToCStringSection<A>::targetCString(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
6132 {
6133 assert(atom->fixupCount() == 1);
6134 ld::Fixup::iterator fit = atom->fixupsBegin();
6135 const ld::Atom* targetAtom = NULL;
6136 switch ( fit->binding ) {
6137 case ld::Fixup::bindingByContentBound:
6138 targetAtom = fit->u.target;
6139 break;
6140 case ld::Fixup::bindingsIndirectlyBound:
6141 targetAtom = ind.indirectAtom(fit->u.bindingIndex);
6142 break;
6143 case ld::Fixup::bindingDirectlyBound:
6144 targetAtom = fit->u.target;
6145 break;
6146 default:
6147 assert(0 && "unsupported reference to selector");
6148 }
6149 assert(targetAtom != NULL);
6150 const Atom<A>* target = dynamic_cast<const Atom<A>*>(targetAtom);
6151 assert(target != NULL);
6152 assert(target->contentType() == ld::Atom::typeCString);
6153 return (char*)target->contentPointer();
6154 }
6155
6156 template <typename A>
6157 unsigned long PointerToCStringSection<A>::contentHash(const class Atom<A>* atom,
6158 const ld::IndirectBindingTable& indirectBindingTable) const
6159 {
6160 // make hash from section name and target cstring name
6161 unsigned long hash = 123;
6162 for (const char* s = this->sectionName(); *s != '\0'; ++s) {
6163 hash = hash * 33 + *s;
6164 }
6165 for (const char* s = this->targetCString(atom, indirectBindingTable); *s != '\0'; ++s) {
6166 hash = hash * 33 + *s;
6167 }
6168 return hash;
6169 }
6170
6171 template <typename A>
6172 bool PointerToCStringSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
6173 const ld::IndirectBindingTable& indirectBindingTable) const
6174 {
6175 assert(this->type() == rhs.section().type());
6176 // there can be pointers-to-cstrings in different section names
6177 // we only want to coalesce in same section name
6178 if ( *this != rhs.section() )
6179 return false;
6180
6181 // get string content for this
6182 const char* cstringContent = this->targetCString(atom, indirectBindingTable);
6183 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
6184 assert(rhsAtom != NULL);
6185 const char* rhsCstringContent = this->targetCString(rhsAtom, indirectBindingTable);
6186
6187 assert(cstringContent != NULL);
6188 assert(rhsCstringContent != NULL);
6189 return (strcmp(cstringContent, rhsCstringContent) == 0);
6190 }
6191
6192
6193
6194 template <typename A>
6195 unsigned long UTF16StringSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
6196 {
6197 unsigned long hash = 5381;
6198 const uint16_t* stringContent = (uint16_t*)atom->contentPointer();
6199 // some buggy compilers end utf16 data with single byte, so don't use last word in hash computation
6200 unsigned int count = (atom->size()/2) - 1;
6201 for (const uint16_t* s = stringContent; count > 0; ++s, --count) {
6202 hash = hash * 33 + *s;
6203 }
6204 return hash;
6205 }
6206
6207 template <typename A>
6208 bool UTF16StringSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
6209 const ld::IndirectBindingTable& ind) const
6210 {
6211 if ( rhs.section().type() != ld::Section::typeUTF16Strings )
6212 return false;
6213 assert(0);
6214 return false;
6215 }
6216
6217
6218
6219
6220
6221
6222
6223 template <>
6224 uint32_t Section<x86_64>::x86_64PcRelOffset(uint8_t r_type)
6225 {
6226 switch ( r_type ) {
6227 case X86_64_RELOC_SIGNED:
6228 return 4;
6229 case X86_64_RELOC_SIGNED_1:
6230 return 5;
6231 case X86_64_RELOC_SIGNED_2:
6232 return 6;
6233 case X86_64_RELOC_SIGNED_4:
6234 return 8;
6235 }
6236 return 0;
6237 }
6238
6239
6240 template <>
6241 bool Section<x86_64>::addRelocFixup(class Parser<x86_64>& parser, const macho_relocation_info<P>* reloc)
6242 {
6243 const macho_section<P>* sect = this->machoSection();
6244 uint64_t srcAddr = sect->addr() + reloc->r_address();
6245 Parser<x86_64>::SourceLocation src;
6246 Parser<x86_64>::TargetDesc target;
6247 Parser<x86_64>::TargetDesc toTarget;
6248 src.atom = this->findAtomByAddress(srcAddr);
6249 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6250 const uint8_t* fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
6251 uint64_t contentValue = 0;
6252 const macho_relocation_info<x86_64::P>* nextReloc = &reloc[1];
6253 bool result = false;
6254 bool useDirectBinding;
6255 switch ( reloc->r_length() ) {
6256 case 0:
6257 contentValue = *fixUpPtr;
6258 break;
6259 case 1:
6260 contentValue = (int64_t)(int16_t)E::get16(*((uint16_t*)fixUpPtr));
6261 break;
6262 case 2:
6263 contentValue = (int64_t)(int32_t)E::get32(*((uint32_t*)fixUpPtr));
6264 break;
6265 case 3:
6266 contentValue = E::get64(*((uint64_t*)fixUpPtr));
6267 break;
6268 }
6269 target.atom = NULL;
6270 target.name = NULL;
6271 target.weakImport = false;
6272 target.addend = 0;
6273 if ( reloc->r_extern() ) {
6274 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
6275 // use direct reference for local symbols
6276 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (parser.nameFromSymbol(sym)[0] == 'L')) ) {
6277 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
6278 target.addend += contentValue;
6279 }
6280 else {
6281 target.name = parser.nameFromSymbol(sym);
6282 target.weakImport = parser.weakImportFromSymbol(sym);
6283 target.addend = contentValue;
6284 }
6285 // cfstrings should always use direct reference to backing store
6286 if ( (this->type() == ld::Section::typeCFString) && (src.offsetInAtom != 0) ) {
6287 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
6288 target.addend = contentValue;
6289 }
6290 }
6291 else {
6292 if ( reloc->r_pcrel() )
6293 contentValue += srcAddr + x86_64PcRelOffset(reloc->r_type());
6294 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), target);
6295 }
6296 switch ( reloc->r_type() ) {
6297 case X86_64_RELOC_UNSIGNED:
6298 if ( reloc->r_pcrel() )
6299 throw "pcrel and X86_64_RELOC_UNSIGNED not supported";
6300 switch ( reloc->r_length() ) {
6301 case 0:
6302 case 1:
6303 throw "length < 2 and X86_64_RELOC_UNSIGNED not supported";
6304 case 2:
6305 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
6306 break;
6307 case 3:
6308 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian64, target);
6309 break;
6310 }
6311 break;
6312 case X86_64_RELOC_SIGNED:
6313 case X86_64_RELOC_SIGNED_1:
6314 case X86_64_RELOC_SIGNED_2:
6315 case X86_64_RELOC_SIGNED_4:
6316 if ( ! reloc->r_pcrel() )
6317 throw "not pcrel and X86_64_RELOC_SIGNED* not supported";
6318 if ( reloc->r_length() != 2 )
6319 throw "length != 2 and X86_64_RELOC_SIGNED* not supported";
6320 switch ( reloc->r_type() ) {
6321 case X86_64_RELOC_SIGNED:
6322 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32, target);
6323 break;
6324 case X86_64_RELOC_SIGNED_1:
6325 if ( reloc->r_extern() )
6326 target.addend += 1;
6327 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32_1, target);
6328 break;
6329 case X86_64_RELOC_SIGNED_2:
6330 if ( reloc->r_extern() )
6331 target.addend += 2;
6332 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32_2, target);
6333 break;
6334 case X86_64_RELOC_SIGNED_4:
6335 if ( reloc->r_extern() )
6336 target.addend += 4;
6337 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32_4, target);
6338 break;
6339 }
6340 break;
6341 case X86_64_RELOC_BRANCH:
6342 if ( ! reloc->r_pcrel() )
6343 throw "not pcrel and X86_64_RELOC_BRANCH not supported";
6344 switch ( reloc->r_length() ) {
6345 case 2:
6346 if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_probe$", 16) == 0) ) {
6347 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreX86DtraceCallSiteNop, false, target.name);
6348 parser.addDtraceExtraInfos(src, &target.name[16]);
6349 }
6350 else if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_isenabled$", 20) == 0) ) {
6351 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreX86DtraceIsEnableSiteClear, false, target.name);
6352 parser.addDtraceExtraInfos(src, &target.name[20]);
6353 }
6354 else {
6355 parser.addFixups(src, ld::Fixup::kindStoreX86BranchPCRel32, target);
6356 }
6357 break;
6358 case 0:
6359 parser.addFixups(src, ld::Fixup::kindStoreX86BranchPCRel8, target);
6360 break;
6361 default:
6362 throwf("length=%d and X86_64_RELOC_BRANCH not supported", reloc->r_length());
6363 }
6364 break;
6365 case X86_64_RELOC_GOT:
6366 if ( ! reloc->r_extern() )
6367 throw "not extern and X86_64_RELOC_GOT not supported";
6368 if ( ! reloc->r_pcrel() )
6369 throw "not pcrel and X86_64_RELOC_GOT not supported";
6370 if ( reloc->r_length() != 2 )
6371 throw "length != 2 and X86_64_RELOC_GOT not supported";
6372 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32GOT, target);
6373 break;
6374 case X86_64_RELOC_GOT_LOAD:
6375 if ( ! reloc->r_extern() )
6376 throw "not extern and X86_64_RELOC_GOT_LOAD not supported";
6377 if ( ! reloc->r_pcrel() )
6378 throw "not pcrel and X86_64_RELOC_GOT_LOAD not supported";
6379 if ( reloc->r_length() != 2 )
6380 throw "length != 2 and X86_64_RELOC_GOT_LOAD not supported";
6381 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32GOTLoad, target);
6382 break;
6383 case X86_64_RELOC_SUBTRACTOR:
6384 if ( reloc->r_pcrel() )
6385 throw "X86_64_RELOC_SUBTRACTOR cannot be pc-relative";
6386 if ( reloc->r_length() < 2 )
6387 throw "X86_64_RELOC_SUBTRACTOR must have r_length of 2 or 3";
6388 if ( !reloc->r_extern() )
6389 throw "X86_64_RELOC_SUBTRACTOR must have r_extern=1";
6390 if ( nextReloc->r_type() != X86_64_RELOC_UNSIGNED )
6391 throw "X86_64_RELOC_SUBTRACTOR must be followed by X86_64_RELOC_UNSIGNED";
6392 result = true;
6393 if ( nextReloc->r_pcrel() )
6394 throw "X86_64_RELOC_UNSIGNED following a X86_64_RELOC_SUBTRACTOR cannot be pc-relative";
6395 if ( nextReloc->r_length() != reloc->r_length() )
6396 throw "X86_64_RELOC_UNSIGNED following a X86_64_RELOC_SUBTRACTOR must have same r_length";
6397 if ( nextReloc->r_extern() ) {
6398 const macho_nlist<P>& sym = parser.symbolFromIndex(nextReloc->r_symbolnum());
6399 // use direct reference for local symbols
6400 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (parser.nameFromSymbol(sym)[0] == 'L')) ) {
6401 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), toTarget);
6402 toTarget.addend = contentValue;
6403 useDirectBinding = true;
6404 }
6405 else {
6406 toTarget.name = parser.nameFromSymbol(sym);
6407 toTarget.weakImport = parser.weakImportFromSymbol(sym);
6408 toTarget.addend = contentValue;
6409 useDirectBinding = false;
6410 }
6411 }
6412 else {
6413 parser.findTargetFromAddressAndSectionNum(contentValue, nextReloc->r_symbolnum(), toTarget);
6414 useDirectBinding = (toTarget.atom->scope() == ld::Atom::scopeTranslationUnit) || ((toTarget.atom->combine() == ld::Atom::combineByNameAndContent) || (toTarget.atom->combine() == ld::Atom::combineByNameAndReferences));
6415 }
6416 if ( useDirectBinding ) {
6417 if ( (toTarget.atom->combine() == ld::Atom::combineByNameAndContent) || (toTarget.atom->combine() == ld::Atom::combineByNameAndReferences) )
6418 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, toTarget.atom);
6419 else
6420 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, toTarget.atom);
6421 }
6422 else
6423 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, toTarget.weakImport, toTarget.name);
6424 parser.addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindAddAddend, toTarget.addend);
6425 if ( target.atom == NULL )
6426 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, false, target.name);
6427 else
6428 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, target.atom);
6429 if ( reloc->r_length() == 2 )
6430 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian32);
6431 else
6432 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian64);
6433 break;
6434 case X86_64_RELOC_TLV:
6435 if ( ! reloc->r_extern() )
6436 throw "not extern and X86_64_RELOC_TLV not supported";
6437 if ( ! reloc->r_pcrel() )
6438 throw "not pcrel and X86_64_RELOC_TLV not supported";
6439 if ( reloc->r_length() != 2 )
6440 throw "length != 2 and X86_64_RELOC_TLV not supported";
6441 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32TLVLoad, target);
6442 break;
6443 default:
6444 throwf("unknown relocation type %d", reloc->r_type());
6445 }
6446 return result;
6447 }
6448
6449
6450
6451 template <>
6452 bool Section<x86>::addRelocFixup(class Parser<x86>& parser, const macho_relocation_info<P>* reloc)
6453 {
6454 const macho_section<P>* sect = this->machoSection();
6455 uint32_t srcAddr;
6456 const uint8_t* fixUpPtr;
6457 uint32_t contentValue = 0;
6458 ld::Fixup::Kind kind = ld::Fixup::kindNone;
6459 Parser<x86>::SourceLocation src;
6460 Parser<x86>::TargetDesc target;
6461
6462 if ( (reloc->r_address() & R_SCATTERED) == 0 ) {
6463 srcAddr = sect->addr() + reloc->r_address();
6464 src.atom = this->findAtomByAddress(srcAddr);
6465 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6466 fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
6467 switch ( reloc->r_type() ) {
6468 case GENERIC_RELOC_VANILLA:
6469 switch ( reloc->r_length() ) {
6470 case 0:
6471 contentValue = (int32_t)(int8_t)*fixUpPtr;
6472 if ( reloc->r_pcrel() ) {
6473 kind = ld::Fixup::kindStoreX86BranchPCRel8;
6474 contentValue += srcAddr + sizeof(uint8_t);
6475 }
6476 else
6477 throw "r_length=0 and r_pcrel=0 not supported";
6478 break;
6479 case 1:
6480 contentValue = (int32_t)(int16_t)E::get16(*((uint16_t*)fixUpPtr));
6481 if ( reloc->r_pcrel() ) {
6482 kind = ld::Fixup::kindStoreX86PCRel16;
6483 contentValue += srcAddr + sizeof(uint16_t);
6484 }
6485 else
6486 kind = ld::Fixup::kindStoreLittleEndian16;
6487 break;
6488 case 2:
6489 contentValue = E::get32(*((uint32_t*)fixUpPtr));
6490 if ( reloc->r_pcrel() ) {
6491 kind = ld::Fixup::kindStoreX86BranchPCRel32;
6492 contentValue += srcAddr + sizeof(uint32_t);
6493 }
6494 else
6495 kind = ld::Fixup::kindStoreLittleEndian32;
6496 break;
6497 case 3:
6498 throw "r_length=3 not supported";
6499 }
6500 if ( reloc->r_extern() ) {
6501 target.atom = NULL;
6502 const macho_nlist<P>& targetSymbol = parser.symbolFromIndex(reloc->r_symbolnum());
6503 target.name = parser.nameFromSymbol(targetSymbol);
6504 target.weakImport = parser.weakImportFromSymbol(targetSymbol);
6505 target.addend = (int32_t)contentValue;
6506 }
6507 else {
6508 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), target);
6509 }
6510 if ( (kind == ld::Fixup::kindStoreX86BranchPCRel32) && (target.name != NULL) ) {
6511 if ( strncmp(target.name, "___dtrace_probe$", 16) == 0 ) {
6512 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreX86DtraceCallSiteNop, false, target.name);
6513 parser.addDtraceExtraInfos(src, &target.name[16]);
6514 return false;
6515 }
6516 else if ( strncmp(target.name, "___dtrace_isenabled$", 20) == 0 ) {
6517 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreX86DtraceIsEnableSiteClear, false, target.name);
6518 parser.addDtraceExtraInfos(src, &target.name[20]);
6519 return false;
6520 }
6521 }
6522 parser.addFixups(src, kind, target);
6523 return false;
6524 break;
6525 case GENERIC_RLEOC_TLV:
6526 {
6527 if ( !reloc->r_extern() )
6528 throw "r_extern=0 and r_type=GENERIC_RLEOC_TLV not supported";
6529 if ( reloc->r_length() != 2 )
6530 throw "r_length!=2 and r_type=GENERIC_RLEOC_TLV not supported";
6531 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
6532 // use direct reference for local symbols
6533 if ( ((sym.n_type() & N_TYPE) == N_SECT) && ((sym.n_type() & N_EXT) == 0) ) {
6534 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
6535 }
6536 else {
6537 target.atom = NULL;
6538 target.name = parser.nameFromSymbol(sym);
6539 target.weakImport = parser.weakImportFromSymbol(sym);
6540 }
6541 target.addend = (int64_t)(int32_t)E::get32(*((uint32_t*)fixUpPtr));
6542 if ( reloc->r_pcrel() ) {
6543 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32TLVLoad, target);
6544 }
6545 else {
6546 parser.addFixups(src, ld::Fixup::kindStoreX86Abs32TLVLoad, target);
6547 }
6548 return false;
6549 }
6550 break;
6551 default:
6552 throwf("unsupported i386 relocation type (%d)", reloc->r_type());
6553 }
6554 }
6555 else {
6556 // scattered relocation
6557 const macho_scattered_relocation_info<P>* sreloc = (macho_scattered_relocation_info<P>*)reloc;
6558 srcAddr = sect->addr() + sreloc->r_address();
6559 src.atom = this->findAtomByAddress(srcAddr);
6560 assert(src.atom != NULL);
6561 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6562 fixUpPtr = file().fileContent() + sect->offset() + sreloc->r_address();
6563 uint32_t relocValue = sreloc->r_value();
6564 bool result = false;
6565 // file format allows pair to be scattered or not
6566 const macho_scattered_relocation_info<P>* nextSReloc = &sreloc[1];
6567 const macho_relocation_info<P>* nextReloc = &reloc[1];
6568 bool nextRelocIsPair = false;
6569 uint32_t nextRelocAddress = 0;
6570 uint32_t nextRelocValue = 0;
6571 if ( (nextReloc->r_address() & R_SCATTERED) == 0 ) {
6572 if ( nextReloc->r_type() == GENERIC_RELOC_PAIR ) {
6573 nextRelocIsPair = true;
6574 nextRelocAddress = nextReloc->r_address();
6575 result = true; // iterator should skip next reloc, since we've consumed it here
6576 }
6577 }
6578 else {
6579 if ( nextSReloc->r_type() == GENERIC_RELOC_PAIR ) {
6580 nextRelocIsPair = true;
6581 nextRelocAddress = nextSReloc->r_address();
6582 nextRelocValue = nextSReloc->r_value();
6583 }
6584 }
6585 switch (sreloc->r_type()) {
6586 case GENERIC_RELOC_VANILLA:
6587 // with a scattered relocation we get both the target (sreloc->r_value()) and the target+offset (*fixUpPtr)
6588 target.atom = parser.findAtomByAddress(relocValue);
6589 if ( sreloc->r_pcrel() ) {
6590 switch ( sreloc->r_length() ) {
6591 case 0:
6592 contentValue = srcAddr + 1 + *fixUpPtr;
6593 target.addend = (int32_t)contentValue - (int32_t)relocValue;
6594 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel8, target);
6595 break;
6596 case 1:
6597 contentValue = srcAddr + 2 + LittleEndian::get16(*((uint16_t*)fixUpPtr));
6598 target.addend = (int32_t)contentValue - (int32_t)relocValue;
6599 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel16, target);
6600 break;
6601 case 2:
6602 contentValue = srcAddr + 4 + LittleEndian::get32(*((uint32_t*)fixUpPtr));
6603 target.addend = (int32_t)contentValue - (int32_t)relocValue;
6604 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32, target);
6605 break;
6606 case 3:
6607 throw "unsupported r_length=3 for scattered pc-rel vanilla reloc";
6608 break;
6609 }
6610 }
6611 else {
6612 if ( sreloc->r_length() != 2 )
6613 throwf("unsupported r_length=%d for scattered vanilla reloc", sreloc->r_length());
6614 contentValue = LittleEndian::get32(*((uint32_t*)fixUpPtr));
6615 target.addend = (int32_t)contentValue - (int32_t)(target.atom->objectAddress());
6616 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
6617 }
6618 break;
6619 case GENERIC_RELOC_SECTDIFF:
6620 case GENERIC_RELOC_LOCAL_SECTDIFF:
6621 {
6622 if ( !nextRelocIsPair )
6623 throw "GENERIC_RELOC_SECTDIFF missing following pair";
6624 switch ( sreloc->r_length() ) {
6625 case 0:
6626 case 3:
6627 throw "bad length for GENERIC_RELOC_SECTDIFF";
6628 case 1:
6629 contentValue = (int32_t)(int16_t)LittleEndian::get16(*((uint16_t*)fixUpPtr));
6630 kind = ld::Fixup::kindStoreLittleEndian16;
6631 break;
6632 case 2:
6633 contentValue = LittleEndian::get32(*((uint32_t*)fixUpPtr));
6634 kind = ld::Fixup::kindStoreLittleEndian32;
6635 break;
6636 }
6637 Atom<x86>* fromAtom = parser.findAtomByAddress(nextRelocValue);
6638 uint32_t offsetInFrom = nextRelocValue - fromAtom->_objAddress;
6639 parser.findTargetFromAddress(sreloc->r_value(), target);
6640 // check for addend encoded in the section content
6641 int64_t addend = (int32_t)contentValue - (int32_t)(sreloc->r_value() - nextRelocValue);
6642 if ( addend < 0 ) {
6643 // switch binding base on coalescing
6644 if ( target.atom == NULL ) {
6645 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, target.name);
6646 }
6647 else if ( target.atom->scope() == ld::Atom::scopeTranslationUnit ) {
6648 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, target.atom);
6649 }
6650 else if ( (target.atom->combine() == ld::Atom::combineByNameAndContent) || (target.atom->combine() == ld::Atom::combineByNameAndReferences) ) {
6651 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, target.atom);
6652 }
6653 else {
6654 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, target.atom->name());
6655 }
6656 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, target.addend);
6657 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
6658 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom-addend);
6659 parser.addFixup(src, ld::Fixup::k5of5, kind);
6660 }
6661 else {
6662 // switch binding base on coalescing
6663 if ( target.atom == NULL ) {
6664 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, target.name);
6665 }
6666 else if ( target.atom->scope() == ld::Atom::scopeTranslationUnit ) {
6667 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, target.atom);
6668 }
6669 else if ( (target.atom->combine() == ld::Atom::combineByNameAndContent) || (target.atom->combine() == ld::Atom::combineByNameAndReferences) ) {
6670 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, target.atom);
6671 }
6672 else {
6673 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, target.atom->name());
6674 }
6675 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, target.addend+addend);
6676 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
6677 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom);
6678 parser.addFixup(src, ld::Fixup::k5of5, kind);
6679 }
6680 }
6681 break;
6682 }
6683 return result;
6684 }
6685 }
6686
6687
6688
6689
6690
6691 #if SUPPORT_ARCH_arm_any
6692 template <>
6693 bool Section<arm>::addRelocFixup(class Parser<arm>& parser, const macho_relocation_info<P>* reloc)
6694 {
6695 const macho_section<P>* sect = this->machoSection();
6696 bool result = false;
6697 uint32_t srcAddr;
6698 uint32_t dstAddr;
6699 uint32_t* fixUpPtr;
6700 int32_t displacement = 0;
6701 uint32_t instruction = 0;
6702 pint_t contentValue = 0;
6703 Parser<arm>::SourceLocation src;
6704 Parser<arm>::TargetDesc target;
6705 const macho_relocation_info<P>* nextReloc;
6706
6707 if ( (reloc->r_address() & R_SCATTERED) == 0 ) {
6708 bool externSymbolIsThumbDef = false;
6709 srcAddr = sect->addr() + reloc->r_address();
6710 src.atom = this->findAtomByAddress(srcAddr);
6711 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6712 fixUpPtr = (uint32_t*)(file().fileContent() + sect->offset() + reloc->r_address());
6713 if ( reloc->r_type() != ARM_RELOC_PAIR )
6714 instruction = LittleEndian::get32(*fixUpPtr);
6715 if ( reloc->r_extern() ) {
6716 const macho_nlist<P>& targetSymbol = parser.symbolFromIndex(reloc->r_symbolnum());
6717 // use direct reference for local symbols
6718 if ( ((targetSymbol.n_type() & N_TYPE) == N_SECT) && (((targetSymbol.n_type() & N_EXT) == 0) || (parser.nameFromSymbol(targetSymbol)[0] == 'L')) ) {
6719 parser.findTargetFromAddressAndSectionNum(targetSymbol.n_value(), targetSymbol.n_sect(), target);
6720 }
6721 else {
6722 target.atom = NULL;
6723 target.name = parser.nameFromSymbol(targetSymbol);
6724 target.weakImport = parser.weakImportFromSymbol(targetSymbol);
6725 if ( ((targetSymbol.n_type() & N_TYPE) == N_SECT) && (targetSymbol.n_desc() & N_ARM_THUMB_DEF) )
6726 externSymbolIsThumbDef = true;
6727 }
6728 }
6729 switch ( reloc->r_type() ) {
6730 case ARM_RELOC_BR24:
6731 // Sign-extend displacement
6732 displacement = (instruction & 0x00FFFFFF) << 2;
6733 if ( (displacement & 0x02000000) != 0 )
6734 displacement |= 0xFC000000;
6735 // The pc added will be +8 from the pc
6736 displacement += 8;
6737 // If this is BLX add H << 1
6738 if ((instruction & 0xFE000000) == 0xFA000000)
6739 displacement += ((instruction & 0x01000000) >> 23);
6740 if ( reloc->r_extern() ) {
6741 dstAddr = srcAddr + displacement;
6742 // <rdar://problem/16652542> support large .o files
6743 if ( srcAddr > 0x2000000 ) {
6744 dstAddr -= ((srcAddr + 0x1FFFFFF) & 0xFC000000);
6745 }
6746 target.addend = dstAddr;
6747 if ( externSymbolIsThumbDef )
6748 target.addend &= -2; // remove thumb bit
6749 }
6750 else {
6751 dstAddr = srcAddr + displacement;
6752 parser.findTargetFromAddressAndSectionNum(dstAddr, reloc->r_symbolnum(), target);
6753 }
6754 // special case "calls" for dtrace
6755 if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_probe$", 16) == 0) ) {
6756 parser.addFixup(src, ld::Fixup::k1of1,
6757 ld::Fixup::kindStoreARMDtraceCallSiteNop, false, target.name);
6758 parser.addDtraceExtraInfos(src, &target.name[16]);
6759 }
6760 else if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_isenabled$", 20) == 0) ) {
6761 parser.addFixup(src, ld::Fixup::k1of1,
6762 ld::Fixup::kindStoreARMDtraceIsEnableSiteClear, false, target.name);
6763 parser.addDtraceExtraInfos(src, &target.name[20]);
6764 }
6765 else {
6766 parser.addFixups(src, ld::Fixup::kindStoreARMBranch24, target);
6767 }
6768 break;
6769 case ARM_THUMB_RELOC_BR22:
6770 // thumb2 added two more bits to displacement, complicating the displacement decoding
6771 {
6772 uint32_t s = (instruction >> 10) & 0x1;
6773 uint32_t j1 = (instruction >> 29) & 0x1;
6774 uint32_t j2 = (instruction >> 27) & 0x1;
6775 uint32_t imm10 = instruction & 0x3FF;
6776 uint32_t imm11 = (instruction >> 16) & 0x7FF;
6777 uint32_t i1 = (j1 == s);
6778 uint32_t i2 = (j2 == s);
6779 uint32_t dis = (s << 24) | (i1 << 23) | (i2 << 22) | (imm10 << 12) | (imm11 << 1);
6780 int32_t sdis = dis;
6781 if ( s )
6782 sdis |= 0xFE000000;
6783 displacement = sdis;
6784 }
6785 // The pc added will be +4 from the pc
6786 displacement += 4;
6787 // If the instruction was blx, force the low 2 bits to be clear
6788 dstAddr = srcAddr + displacement;
6789 if ((instruction & 0xD0000000) == 0xC0000000)
6790 dstAddr &= 0xFFFFFFFC;
6791
6792 if ( reloc->r_extern() ) {
6793 // <rdar://problem/16652542> support large .o files
6794 if ( srcAddr > 0x1000000 ) {
6795 dstAddr -= ((srcAddr + 0xFFFFFF) & 0xFE000000);
6796 }
6797 target.addend = (int64_t)(int32_t)dstAddr;
6798 }
6799 else {
6800 parser.findTargetFromAddressAndSectionNum(dstAddr, reloc->r_symbolnum(), target);
6801 }
6802 // special case "calls" for dtrace
6803 if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_probe$", 16) == 0) ) {
6804 parser.addFixup(src, ld::Fixup::k1of1,
6805 ld::Fixup::kindStoreThumbDtraceCallSiteNop, false, target.name);
6806 parser.addDtraceExtraInfos(src, &target.name[16]);
6807 }
6808 else if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_isenabled$", 20) == 0) ) {
6809 parser.addFixup(src, ld::Fixup::k1of1,
6810 ld::Fixup::kindStoreThumbDtraceIsEnableSiteClear, false, target.name);
6811 parser.addDtraceExtraInfos(src, &target.name[20]);
6812 }
6813 else {
6814 parser.addFixups(src, ld::Fixup::kindStoreThumbBranch22, target);
6815 }
6816 break;
6817 case ARM_RELOC_VANILLA:
6818 if ( reloc->r_length() != 2 )
6819 throw "bad length for ARM_RELOC_VANILLA";
6820 contentValue = LittleEndian::get32(*fixUpPtr);
6821 if ( reloc->r_extern() ) {
6822 target.addend = (int32_t)contentValue;
6823 if ( externSymbolIsThumbDef )
6824 target.addend &= -2; // remove thumb bit
6825 }
6826 else {
6827 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), target);
6828 // possible non-extern relocation turned into by-name ref because target is a weak-def
6829 if ( target.atom != NULL ) {
6830 if ( target.atom->isThumb() )
6831 target.addend &= -2; // remove thumb bit
6832 // if reference to LSDA, add group subordinate fixup
6833 if ( target.atom->contentType() == ld::Atom::typeLSDA ) {
6834 Parser<arm>::SourceLocation src2;
6835 src2.atom = src.atom;
6836 src2.offsetInAtom = 0;
6837 parser.addFixup(src2, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateLSDA, target.atom);
6838 }
6839 }
6840 }
6841 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
6842 break;
6843 case ARM_THUMB_32BIT_BRANCH:
6844 // silently ignore old unnecessary reloc
6845 break;
6846 case ARM_RELOC_HALF:
6847 nextReloc = &reloc[1];
6848 if ( nextReloc->r_type() == ARM_RELOC_PAIR ) {
6849 uint32_t instruction16;
6850 uint32_t other16 = (nextReloc->r_address() & 0xFFFF);
6851 bool isThumb;
6852 if ( reloc->r_length() & 2 ) {
6853 isThumb = true;
6854 uint32_t i = ((instruction & 0x00000400) >> 10);
6855 uint32_t imm4 = (instruction & 0x0000000F);
6856 uint32_t imm3 = ((instruction & 0x70000000) >> 28);
6857 uint32_t imm8 = ((instruction & 0x00FF0000) >> 16);
6858 instruction16 = (imm4 << 12) | (i << 11) | (imm3 << 8) | imm8;
6859 }
6860 else {
6861 isThumb = false;
6862 uint32_t imm4 = ((instruction & 0x000F0000) >> 16);
6863 uint32_t imm12 = (instruction & 0x00000FFF);
6864 instruction16 = (imm4 << 12) | imm12;
6865 }
6866 if ( reloc->r_length() & 1 ) {
6867 // high 16
6868 dstAddr = ((instruction16 << 16) | other16);
6869 if ( reloc->r_extern() ) {
6870 target.addend = dstAddr;
6871 if ( externSymbolIsThumbDef )
6872 target.addend &= -2; // remove thumb bit
6873 }
6874 else {
6875 parser.findTargetFromAddress(dstAddr, target);
6876 if ( target.atom->isThumb() )
6877 target.addend &= (-2); // remove thumb bit
6878 }
6879 parser.addFixups(src, (isThumb ? ld::Fixup::kindStoreThumbHigh16 : ld::Fixup::kindStoreARMHigh16), target);
6880 }
6881 else {
6882 // low 16
6883 dstAddr = (other16 << 16) | instruction16;
6884 if ( reloc->r_extern() ) {
6885 target.addend = dstAddr;
6886 if ( externSymbolIsThumbDef )
6887 target.addend &= -2; // remove thumb bit
6888 }
6889 else {
6890 parser.findTargetFromAddress(dstAddr, target);
6891 if ( target.atom->isThumb() )
6892 target.addend &= (-2); // remove thumb bit
6893 }
6894 parser.addFixups(src, (isThumb ? ld::Fixup::kindStoreThumbLow16 : ld::Fixup::kindStoreARMLow16), target);
6895 }
6896 result = true;
6897 }
6898 else
6899 throw "for ARM_RELOC_HALF, next reloc is not ARM_RELOC_PAIR";
6900 break;
6901 default:
6902 throwf("unknown relocation type %d", reloc->r_type());
6903 break;
6904 }
6905 }
6906 else {
6907 const macho_scattered_relocation_info<P>* sreloc = (macho_scattered_relocation_info<P>*)reloc;
6908 // file format allows pair to be scattered or not
6909 const macho_scattered_relocation_info<P>* nextSReloc = &sreloc[1];
6910 nextReloc = &reloc[1];
6911 srcAddr = sect->addr() + sreloc->r_address();
6912 dstAddr = sreloc->r_value();
6913 fixUpPtr = (uint32_t*)(file().fileContent() + sect->offset() + sreloc->r_address());
6914 instruction = LittleEndian::get32(*fixUpPtr);
6915 src.atom = this->findAtomByAddress(srcAddr);
6916 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6917 bool nextRelocIsPair = false;
6918 uint32_t nextRelocAddress = 0;
6919 uint32_t nextRelocValue = 0;
6920 if ( (nextReloc->r_address() & R_SCATTERED) == 0 ) {
6921 if ( nextReloc->r_type() == ARM_RELOC_PAIR ) {
6922 nextRelocIsPair = true;
6923 nextRelocAddress = nextReloc->r_address();
6924 result = true;
6925 }
6926 }
6927 else {
6928 if ( nextSReloc->r_type() == ARM_RELOC_PAIR ) {
6929 nextRelocIsPair = true;
6930 nextRelocAddress = nextSReloc->r_address();
6931 nextRelocValue = nextSReloc->r_value();
6932 result = true;
6933 }
6934 }
6935 switch ( sreloc->r_type() ) {
6936 case ARM_RELOC_VANILLA:
6937 // with a scattered relocation we get both the target (sreloc->r_value()) and the target+offset (*fixUpPtr)
6938 if ( sreloc->r_length() != 2 )
6939 throw "bad length for ARM_RELOC_VANILLA";
6940 target.atom = parser.findAtomByAddress(sreloc->r_value());
6941 if ( target.atom == NULL )
6942 throwf("bad r_value (0x%08X) for ARM_RELOC_VANILLA\n", sreloc->r_value());
6943 contentValue = LittleEndian::get32(*fixUpPtr);
6944 target.addend = contentValue - target.atom->_objAddress;
6945 if ( target.atom->isThumb() )
6946 target.addend &= -2; // remove thumb bit
6947 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
6948 break;
6949 case ARM_RELOC_BR24:
6950 // Sign-extend displacement
6951 displacement = (instruction & 0x00FFFFFF) << 2;
6952 if ( (displacement & 0x02000000) != 0 )
6953 displacement |= 0xFC000000;
6954 // The pc added will be +8 from the pc
6955 displacement += 8;
6956 // If this is BLX add H << 1
6957 if ((instruction & 0xFE000000) == 0xFA000000)
6958 displacement += ((instruction & 0x01000000) >> 23);
6959 target.atom = parser.findAtomByAddress(sreloc->r_value());
6960 target.addend = (int64_t)(srcAddr + displacement) - (int64_t)(target.atom->_objAddress);
6961 parser.addFixups(src, ld::Fixup::kindStoreARMBranch24, target);
6962 break;
6963 case ARM_THUMB_RELOC_BR22:
6964 // thumb2 added two more bits to displacement, complicating the displacement decoding
6965 {
6966 uint32_t s = (instruction >> 10) & 0x1;
6967 uint32_t j1 = (instruction >> 29) & 0x1;
6968 uint32_t j2 = (instruction >> 27) & 0x1;
6969 uint32_t imm10 = instruction & 0x3FF;
6970 uint32_t imm11 = (instruction >> 16) & 0x7FF;
6971 uint32_t i1 = (j1 == s);
6972 uint32_t i2 = (j2 == s);
6973 uint32_t dis = (s << 24) | (i1 << 23) | (i2 << 22) | (imm10 << 12) | (imm11 << 1);
6974 int32_t sdis = dis;
6975 if ( s )
6976 sdis |= 0xFE000000;
6977 displacement = sdis;
6978 }
6979 // The pc added will be +4 from the pc
6980 displacement += 4;
6981 dstAddr = srcAddr+displacement;
6982 // If the instruction was blx, force the low 2 bits to be clear
6983 if ((instruction & 0xF8000000) == 0xE8000000)
6984 dstAddr &= 0xFFFFFFFC;
6985 target.atom = parser.findAtomByAddress(sreloc->r_value());
6986 target.addend = dstAddr - target.atom->_objAddress;
6987 parser.addFixups(src, ld::Fixup::kindStoreThumbBranch22, target);
6988 break;
6989 case ARM_RELOC_SECTDIFF:
6990 case ARM_RELOC_LOCAL_SECTDIFF:
6991 {
6992 if ( ! nextRelocIsPair )
6993 throw "ARM_RELOC_SECTDIFF missing following pair";
6994 if ( sreloc->r_length() != 2 )
6995 throw "bad length for ARM_RELOC_SECTDIFF";
6996 contentValue = LittleEndian::get32(*fixUpPtr);
6997 Atom<arm>* fromAtom = parser.findAtomByAddress(nextRelocValue);
6998 uint32_t offsetInFrom = nextRelocValue - fromAtom->_objAddress;
6999 uint32_t offsetInTarget;
7000 Atom<arm>* targetAtom = parser.findAtomByAddressOrLocalTargetOfStub(sreloc->r_value(), &offsetInTarget);
7001 // check for addend encoded in the section content
7002 int64_t addend = (int32_t)contentValue - (int32_t)(sreloc->r_value() - nextRelocValue);
7003 if ( targetAtom->isThumb() )
7004 addend &= -2; // remove thumb bit
7005 // if reference to LSDA, add group subordinate fixup
7006 if ( targetAtom->contentType() == ld::Atom::typeLSDA ) {
7007 Parser<arm>::SourceLocation src2;
7008 src2.atom = src.atom;
7009 src2.offsetInAtom = 0;
7010 parser.addFixup(src2, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateLSDA, targetAtom);
7011 }
7012 if ( addend < 0 ) {
7013 // switch binding base on coalescing
7014 if ( targetAtom->scope() == ld::Atom::scopeTranslationUnit ) {
7015 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, targetAtom);
7016 }
7017 else if ( (targetAtom->combine() == ld::Atom::combineByNameAndContent) || (targetAtom->combine() == ld::Atom::combineByNameAndReferences) ) {
7018 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, targetAtom);
7019 }
7020 else {
7021 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, targetAtom->name());
7022 }
7023 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, offsetInTarget);
7024 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
7025 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom-addend);
7026 parser.addFixup(src, ld::Fixup::k5of5, ld::Fixup::kindStoreLittleEndian32);
7027 }
7028 else {
7029 if ( targetAtom->scope() == ld::Atom::scopeTranslationUnit ) {
7030 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, targetAtom);
7031 }
7032 else if ( (targetAtom->combine() == ld::Atom::combineByNameAndContent) || (targetAtom->combine() == ld::Atom::combineByNameAndReferences) ) {
7033 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, targetAtom);
7034 }
7035 else {
7036 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, targetAtom->name());
7037 }
7038 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, (uint32_t)(offsetInTarget+addend));
7039 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
7040 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom);
7041 parser.addFixup(src, ld::Fixup::k5of5, ld::Fixup::kindStoreLittleEndian32);
7042 }
7043 }
7044 break;
7045 case ARM_RELOC_HALF_SECTDIFF:
7046 if ( nextRelocIsPair ) {
7047 instruction = LittleEndian::get32(*fixUpPtr);
7048 Atom<arm>* fromAtom = parser.findAtomByAddress(nextRelocValue);
7049 uint32_t offsetInFrom = nextRelocValue - fromAtom->_objAddress;
7050 Atom<arm>* targetAtom = parser.findAtomByAddress(sreloc->r_value());
7051 uint32_t offsetInTarget = sreloc->r_value() - targetAtom->_objAddress;
7052 uint32_t instruction16;
7053 uint32_t other16 = (nextRelocAddress & 0xFFFF);
7054 bool isThumb;
7055 if ( sreloc->r_length() & 2 ) {
7056 isThumb = true;
7057 uint32_t i = ((instruction & 0x00000400) >> 10);
7058 uint32_t imm4 = (instruction & 0x0000000F);
7059 uint32_t imm3 = ((instruction & 0x70000000) >> 28);
7060 uint32_t imm8 = ((instruction & 0x00FF0000) >> 16);
7061 instruction16 = (imm4 << 12) | (i << 11) | (imm3 << 8) | imm8;
7062 }
7063 else {
7064 isThumb = false;
7065 uint32_t imm4 = ((instruction & 0x000F0000) >> 16);
7066 uint32_t imm12 = (instruction & 0x00000FFF);
7067 instruction16 = (imm4 << 12) | imm12;
7068 }
7069 if ( sreloc->r_length() & 1 )
7070 dstAddr = ((instruction16 << 16) | other16);
7071 else
7072 dstAddr = (other16 << 16) | instruction16;
7073 if ( targetAtom->isThumb() )
7074 dstAddr &= (-2); // remove thumb bit
7075 int32_t addend = dstAddr - (sreloc->r_value() - nextRelocValue);
7076 if ( targetAtom->scope() == ld::Atom::scopeTranslationUnit ) {
7077 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, targetAtom);
7078 }
7079 else if ( (targetAtom->combine() == ld::Atom::combineByNameAndContent) || (targetAtom->combine() == ld::Atom::combineByNameAndReferences) ) {
7080 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, targetAtom);
7081 }
7082 else {
7083 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, targetAtom->name());
7084 }
7085 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, (uint32_t)offsetInTarget+addend);
7086 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
7087 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom);
7088 if ( sreloc->r_length() & 1 ) {
7089 // high 16
7090 parser.addFixup(src, ld::Fixup::k5of5, (isThumb ? ld::Fixup::kindStoreThumbHigh16 : ld::Fixup::kindStoreARMHigh16));
7091 }
7092 else {
7093 // low 16
7094 parser.addFixup(src, ld::Fixup::k5of5, (isThumb ? ld::Fixup::kindStoreThumbLow16 : ld::Fixup::kindStoreARMLow16));
7095 }
7096 result = true;
7097 }
7098 else
7099 throw "ARM_RELOC_HALF_SECTDIFF reloc missing following pair";
7100 break;
7101 case ARM_RELOC_HALF:
7102 if ( nextRelocIsPair ) {
7103 instruction = LittleEndian::get32(*fixUpPtr);
7104 Atom<arm>* targetAtom = parser.findAtomByAddress(sreloc->r_value());
7105 uint32_t instruction16;
7106 uint32_t other16 = (nextRelocAddress & 0xFFFF);
7107 bool isThumb;
7108 if ( sreloc->r_length() & 2 ) {
7109 isThumb = true;
7110 uint32_t i = ((instruction & 0x00000400) >> 10);
7111 uint32_t imm4 = (instruction & 0x0000000F);
7112 uint32_t imm3 = ((instruction & 0x70000000) >> 28);
7113 uint32_t imm8 = ((instruction & 0x00FF0000) >> 16);
7114 instruction16 = (imm4 << 12) | (i << 11) | (imm3 << 8) | imm8;
7115 }
7116 else {
7117 isThumb = false;
7118 uint32_t imm4 = ((instruction & 0x000F0000) >> 16);
7119 uint32_t imm12 = (instruction & 0x00000FFF);
7120 instruction16 = (imm4 << 12) | imm12;
7121 }
7122 if ( sreloc->r_length() & 1 )
7123 dstAddr = ((instruction16 << 16) | other16);
7124 else
7125 dstAddr = (other16 << 16) | instruction16;
7126 if ( targetAtom->scope() == ld::Atom::scopeTranslationUnit ) {
7127 parser.addFixup(src, ld::Fixup::k1of3, ld::Fixup::kindSetTargetAddress, targetAtom);
7128 }
7129 else if ( (targetAtom->combine() == ld::Atom::combineByNameAndContent) || (targetAtom->combine() == ld::Atom::combineByNameAndReferences) ) {
7130 parser.addFixup(src, ld::Fixup::k1of3, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, targetAtom);
7131 }
7132 else {
7133 parser.addFixup(src, ld::Fixup::k1of3, ld::Fixup::kindSetTargetAddress, false, targetAtom->name());
7134 }
7135 parser.addFixup(src, ld::Fixup::k2of3, ld::Fixup::kindAddAddend, dstAddr - targetAtom->_objAddress);
7136 if ( sreloc->r_length() & 1 ) {
7137 // high 16
7138 parser.addFixup(src, ld::Fixup::k3of3, (isThumb ? ld::Fixup::kindStoreThumbHigh16 : ld::Fixup::kindStoreARMHigh16));
7139 }
7140 else {
7141 // low 16
7142 parser.addFixup(src, ld::Fixup::k3of3, (isThumb ? ld::Fixup::kindStoreThumbLow16 : ld::Fixup::kindStoreARMLow16));
7143 }
7144 result = true;
7145 }
7146 else
7147 throw "scattered ARM_RELOC_HALF reloc missing following pair";
7148 break;
7149 default:
7150 throwf("unknown ARM scattered relocation type %d", sreloc->r_type());
7151 }
7152 }
7153 return result;
7154 }
7155 #endif
7156
7157
7158 #if SUPPORT_ARCH_arm64
7159 template <>
7160 bool Section<arm64>::addRelocFixup(class Parser<arm64>& parser, const macho_relocation_info<P>* reloc)
7161 {
7162 bool result = false;
7163 Parser<arm64>::SourceLocation src;
7164 Parser<arm64>::TargetDesc target = { NULL, NULL, false, 0 };
7165 Parser<arm64>::TargetDesc toTarget;
7166 int32_t prefixRelocAddend = 0;
7167 if ( reloc->r_type() == ARM64_RELOC_ADDEND ) {
7168 uint32_t rawAddend = reloc->r_symbolnum();
7169 prefixRelocAddend = rawAddend;
7170 if ( rawAddend & 0x00800000 )
7171 prefixRelocAddend |= 0xFF000000; // sign extend 24-bit signed int to 32-bits
7172 uint32_t addendAddress = reloc->r_address();
7173 ++reloc; //advance to next reloc record
7174 result = true;
7175 if ( reloc->r_address() != addendAddress )
7176 throw "ARM64_RELOC_ADDEND r_address does not match next reloc's r_address";
7177 }
7178 const macho_section<P>* sect = this->machoSection();
7179 uint64_t srcAddr = sect->addr() + reloc->r_address();
7180 src.atom = this->findAtomByAddress(srcAddr);
7181 src.offsetInAtom = srcAddr - src.atom->_objAddress;
7182 const uint8_t* fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
7183 uint64_t contentValue = 0;
7184 const macho_relocation_info<arm64::P>* nextReloc = &reloc[1];
7185 bool useDirectBinding;
7186 uint32_t instruction;
7187 uint32_t encodedAddend;
7188 switch ( reloc->r_length() ) {
7189 case 0:
7190 contentValue = *fixUpPtr;
7191 break;
7192 case 1:
7193 contentValue = (int64_t)(int16_t)E::get16(*((uint16_t*)fixUpPtr));
7194 break;
7195 case 2:
7196 contentValue = (int64_t)(int32_t)E::get32(*((uint32_t*)fixUpPtr));
7197 break;
7198 case 3:
7199 contentValue = E::get64(*((uint64_t*)fixUpPtr));
7200 break;
7201 }
7202 if ( reloc->r_extern() ) {
7203 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
7204 const char* symbolName = parser.nameFromSymbol(sym);
7205 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (symbolName[0] == 'L') || (symbolName[0] == 'l')) ) {
7206 // use direct reference for local symbols
7207 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
7208 //target.addend += contentValue;
7209 }
7210 else if ( ((sym.n_type() & N_TYPE) == N_SECT) && (src.atom->_objAddress <= sym.n_value()) && (sym.n_value() < (src.atom->_objAddress+src.atom->size())) ) {
7211 // <rdar://problem/13700961> spurious warning when weak function has reference to itself
7212 // use direct reference when atom targets itself
7213 target.atom = src.atom;
7214 target.name = NULL;
7215 }
7216 else {
7217 target.name = symbolName;
7218 target.weakImport = parser.weakImportFromSymbol(sym);
7219 //target.addend = contentValue;
7220 }
7221 // cfstrings should always use direct reference to backing store
7222 if ( (this->type() == ld::Section::typeCFString) && (src.offsetInAtom != 0) ) {
7223 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
7224 //target.addend = contentValue;
7225 }
7226 }
7227 else {
7228 if ( reloc->r_pcrel() )
7229 contentValue += srcAddr;
7230 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), target);
7231 }
7232 switch ( reloc->r_type() ) {
7233 case ARM64_RELOC_UNSIGNED:
7234 if ( reloc->r_pcrel() )
7235 throw "pcrel and ARM64_RELOC_UNSIGNED not supported";
7236 if ( reloc->r_extern() )
7237 target.addend = contentValue;
7238 switch ( reloc->r_length() ) {
7239 case 0:
7240 case 1:
7241 throw "length < 2 and ARM64_RELOC_UNSIGNED not supported";
7242 case 2:
7243 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
7244 break;
7245 case 3:
7246 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian64, target);
7247 break;
7248 }
7249 break;
7250 case ARM64_RELOC_BRANCH26:
7251 if ( ! reloc->r_pcrel() )
7252 throw "not pcrel and ARM64_RELOC_BRANCH26 not supported";
7253 if ( ! reloc->r_extern() )
7254 throw "r_extern == 0 and ARM64_RELOC_BRANCH26 not supported";
7255 if ( reloc->r_length() != 2 )
7256 throw "r_length != 2 and ARM64_RELOC_BRANCH26 not supported";
7257 if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_probe$", 16) == 0) ) {
7258 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreARM64DtraceCallSiteNop, false, target.name);
7259 parser.addDtraceExtraInfos(src, &target.name[16]);
7260 }
7261 else if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_isenabled$", 20) == 0) ) {
7262 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreARM64DtraceIsEnableSiteClear, false, target.name);
7263 parser.addDtraceExtraInfos(src, &target.name[20]);
7264 }
7265 else {
7266 target.addend = prefixRelocAddend;
7267 instruction = contentValue;
7268 encodedAddend = (instruction & 0x03FFFFFF) << 2;
7269 if ( encodedAddend != 0 ) {
7270 if ( prefixRelocAddend == 0 ) {
7271 warning("branch26 instruction at 0x%08X has embedded addend. ARM64_RELOC_ADDEND should be used instead", reloc->r_address());
7272 target.addend = encodedAddend;
7273 }
7274 else {
7275 throwf("branch26 instruction at 0x%08X has embedded addend and ARM64_RELOC_ADDEND also used", reloc->r_address());
7276 }
7277 }
7278 parser.addFixups(src, ld::Fixup::kindStoreARM64Branch26, target);
7279 }
7280 break;
7281 case ARM64_RELOC_PAGE21:
7282 if ( ! reloc->r_pcrel() )
7283 throw "not pcrel and ARM64_RELOC_PAGE21 not supported";
7284 if ( ! reloc->r_extern() )
7285 throw "r_extern == 0 and ARM64_RELOC_PAGE21 not supported";
7286 if ( reloc->r_length() != 2 )
7287 throw "length != 2 and ARM64_RELOC_PAGE21 not supported";
7288 target.addend = prefixRelocAddend;
7289 instruction = contentValue;
7290 encodedAddend = ((instruction & 0x60000000) >> 29) | ((instruction & 0x01FFFFE0) >> 3);
7291 encodedAddend *= 4096; // internally addend is in bytes, so scale
7292 if ( encodedAddend != 0 ) {
7293 if ( prefixRelocAddend == 0 ) {
7294 warning("adrp instruction at 0x%08X has embedded addend. ARM64_RELOC_ADDEND should be used instead", reloc->r_address());
7295 target.addend = encodedAddend;
7296 }
7297 else {
7298 throwf("adrp instruction at 0x%08X has embedded addend and ARM64_RELOC_ADDEND also used", reloc->r_address());
7299 }
7300 }
7301 parser.addFixups(src, ld::Fixup::kindStoreARM64Page21, target);
7302 break;
7303 case ARM64_RELOC_PAGEOFF12:
7304 if ( reloc->r_pcrel() )
7305 throw "pcrel and ARM64_RELOC_PAGEOFF12 not supported";
7306 if ( ! reloc->r_extern() )
7307 throw "r_extern == 0 and ARM64_RELOC_PAGEOFF12 not supported";
7308 if ( reloc->r_length() != 2 )
7309 throw "length != 2 and ARM64_RELOC_PAGEOFF12 not supported";
7310 target.addend = prefixRelocAddend;
7311 instruction = contentValue;
7312 encodedAddend = ((instruction & 0x003FFC00) >> 10);
7313 // internally addend is in bytes. Some instructions have an implicit scale factor
7314 if ( (instruction & 0x3B000000) == 0x39000000 ) {
7315 switch ( instruction & 0xC0000000 ) {
7316 case 0x00000000:
7317 break;
7318 case 0x40000000:
7319 encodedAddend *= 2;
7320 break;
7321 case 0x80000000:
7322 encodedAddend *= 4;
7323 break;
7324 case 0xC0000000:
7325 encodedAddend *= 8;
7326 break;
7327 }
7328 }
7329 if ( encodedAddend != 0 ) {
7330 if ( prefixRelocAddend == 0 ) {
7331 warning("pageoff12 instruction at 0x%08X has embedded addend. ARM64_RELOC_ADDEND should be used instead", reloc->r_address());
7332 target.addend = encodedAddend;
7333 }
7334 else {
7335 throwf("pageoff12 instruction at 0x%08X has embedded addend and ARM64_RELOC_ADDEND also used", reloc->r_address());
7336 }
7337 }
7338 parser.addFixups(src, ld::Fixup::kindStoreARM64PageOff12, target);
7339 break;
7340 case ARM64_RELOC_GOT_LOAD_PAGE21:
7341 if ( ! reloc->r_pcrel() )
7342 throw "not pcrel and ARM64_RELOC_GOT_LOAD_PAGE21 not supported";
7343 if ( ! reloc->r_extern() )
7344 throw "r_extern == 0 and ARM64_RELOC_GOT_LOAD_PAGE21 not supported";
7345 if ( reloc->r_length() != 2 )
7346 throw "length != 2 and ARM64_RELOC_GOT_LOAD_PAGE21 not supported";
7347 if ( prefixRelocAddend != 0 )
7348 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_GOT_LOAD_PAGE21 not supported";
7349 instruction = contentValue;
7350 target.addend = ((instruction & 0x60000000) >> 29) | ((instruction & 0x01FFFFE0) >> 3);
7351 if ( target.addend != 0 )
7352 throw "non-zero addend with ARM64_RELOC_GOT_LOAD_PAGE21 is not supported";
7353 parser.addFixups(src, ld::Fixup::kindStoreARM64GOTLoadPage21, target);
7354 break;
7355 case ARM64_RELOC_GOT_LOAD_PAGEOFF12:
7356 if ( reloc->r_pcrel() )
7357 throw "pcrel and ARM64_RELOC_GOT_LOAD_PAGEOFF12 not supported";
7358 if ( ! reloc->r_extern() )
7359 throw "r_extern == 0 and ARM64_RELOC_GOT_LOAD_PAGEOFF12 not supported";
7360 if ( reloc->r_length() != 2 )
7361 throw "length != 2 and ARM64_RELOC_GOT_LOAD_PAGEOFF12 not supported";
7362 if ( prefixRelocAddend != 0 )
7363 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_GOT_LOAD_PAGEOFF12 not supported";
7364 instruction = contentValue;
7365 target.addend = ((instruction & 0x003FFC00) >> 10);
7366 parser.addFixups(src, ld::Fixup::kindStoreARM64GOTLoadPageOff12, target);
7367 break;
7368 case ARM64_RELOC_TLVP_LOAD_PAGE21:
7369 if ( ! reloc->r_pcrel() )
7370 throw "not pcrel and ARM64_RELOC_TLVP_LOAD_PAGE21 not supported";
7371 if ( ! reloc->r_extern() )
7372 throw "r_extern == 0 and ARM64_RELOC_TLVP_LOAD_PAGE21 not supported";
7373 if ( reloc->r_length() != 2 )
7374 throw "length != 2 and ARM64_RELOC_TLVP_LOAD_PAGE21 not supported";
7375 if ( prefixRelocAddend != 0 )
7376 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_TLVP_LOAD_PAGE21 not supported";
7377 instruction = contentValue;
7378 target.addend = ((instruction & 0x60000000) >> 29) | ((instruction & 0x01FFFFE0) >> 3);
7379 if ( target.addend != 0 )
7380 throw "non-zero addend with ARM64_RELOC_GOT_LOAD_PAGE21 is not supported";
7381 parser.addFixups(src, ld::Fixup::kindStoreARM64TLVPLoadPage21, target);
7382 break;
7383 case ARM64_RELOC_TLVP_LOAD_PAGEOFF12:
7384 if ( reloc->r_pcrel() )
7385 throw "pcrel and ARM64_RELOC_TLVP_LOAD_PAGEOFF12 not supported";
7386 if ( ! reloc->r_extern() )
7387 throw "r_extern == 0 and ARM64_RELOC_TLVP_LOAD_PAGEOFF12 not supported";
7388 if ( reloc->r_length() != 2 )
7389 throw "length != 2 and ARM64_RELOC_TLVP_LOAD_PAGEOFF12 not supported";
7390 if ( prefixRelocAddend != 0 )
7391 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_TLVP_LOAD_PAGEOFF12 not supported";
7392 instruction = contentValue;
7393 target.addend = ((instruction & 0x003FFC00) >> 10);
7394 parser.addFixups(src, ld::Fixup::kindStoreARM64TLVPLoadPageOff12, target);
7395 break;
7396 case ARM64_RELOC_SUBTRACTOR:
7397 if ( reloc->r_pcrel() )
7398 throw "ARM64_RELOC_SUBTRACTOR cannot be pc-relative";
7399 if ( reloc->r_length() < 2 )
7400 throw "ARM64_RELOC_SUBTRACTOR must have r_length of 2 or 3";
7401 if ( !reloc->r_extern() )
7402 throw "ARM64_RELOC_SUBTRACTOR must have r_extern=1";
7403 if ( nextReloc->r_type() != ARM64_RELOC_UNSIGNED )
7404 throw "ARM64_RELOC_SUBTRACTOR must be followed by ARM64_RELOC_UNSIGNED";
7405 if ( prefixRelocAddend != 0 )
7406 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_SUBTRACTOR not supported";
7407 result = true;
7408 if ( nextReloc->r_pcrel() )
7409 throw "ARM64_RELOC_UNSIGNED following a ARM64_RELOC_SUBTRACTOR cannot be pc-relative";
7410 if ( nextReloc->r_length() != reloc->r_length() )
7411 throw "ARM64_RELOC_UNSIGNED following a ARM64_RELOC_SUBTRACTOR must have same r_length";
7412 if ( nextReloc->r_extern() ) {
7413 const macho_nlist<P>& sym = parser.symbolFromIndex(nextReloc->r_symbolnum());
7414 // use direct reference for local symbols
7415 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (parser.nameFromSymbol(sym)[0] == 'L')) ) {
7416 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), toTarget);
7417 toTarget.addend = contentValue;
7418 useDirectBinding = true;
7419 }
7420 else {
7421 toTarget.name = parser.nameFromSymbol(sym);
7422 toTarget.weakImport = parser.weakImportFromSymbol(sym);
7423 toTarget.addend = contentValue;
7424 useDirectBinding = false;
7425 }
7426 }
7427 else {
7428 parser.findTargetFromAddressAndSectionNum(contentValue, nextReloc->r_symbolnum(), toTarget);
7429 useDirectBinding = (toTarget.atom->scope() == ld::Atom::scopeTranslationUnit);
7430 }
7431 if ( useDirectBinding )
7432 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, toTarget.atom);
7433 else
7434 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, toTarget.weakImport, toTarget.name);
7435 parser.addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindAddAddend, toTarget.addend);
7436 if ( target.atom == NULL )
7437 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, false, target.name);
7438 else
7439 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, target.atom);
7440 if ( reloc->r_length() == 2 )
7441 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian32);
7442 else
7443 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian64);
7444 break;
7445 case ARM64_RELOC_POINTER_TO_GOT:
7446 if ( ! reloc->r_extern() )
7447 throw "r_extern == 0 and ARM64_RELOC_POINTER_TO_GOT not supported";
7448 if ( prefixRelocAddend != 0 )
7449 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_POINTER_TO_GOT not supported";
7450 if ( reloc->r_pcrel() ) {
7451 if ( reloc->r_length() != 2 )
7452 throw "r_length != 2 and r_extern = 1 and ARM64_RELOC_POINTER_TO_GOT not supported";
7453 parser.addFixups(src, ld::Fixup::kindStoreARM64PCRelToGOT, target);
7454 }
7455 else {
7456 if ( reloc->r_length() != 3 )
7457 throw "r_length != 3 and r_extern = 0 and ARM64_RELOC_POINTER_TO_GOT not supported";
7458 parser.addFixups(src, ld::Fixup::kindStoreARM64PointerToGOT, target);
7459 }
7460 break;
7461 default:
7462 throwf("unknown relocation type %d", reloc->r_type());
7463 }
7464 return result;
7465 }
7466 #endif
7467
7468
7469 template <typename A>
7470 bool ObjC1ClassSection<A>::addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>* reloc)
7471 {
7472 // inherited
7473 FixedSizeSection<A>::addRelocFixup(parser, reloc);
7474
7475 assert(0 && "needs template specialization");
7476 return false;
7477 }
7478
7479 template <>
7480 bool ObjC1ClassSection<x86>::addRelocFixup(class Parser<x86>& parser, const macho_relocation_info<x86::P>* reloc)
7481 {
7482 // if this is the reloc for the super class name string, add implicit reference to super class
7483 if ( ((reloc->r_address() & R_SCATTERED) == 0) && (reloc->r_type() == GENERIC_RELOC_VANILLA) ) {
7484 assert( reloc->r_length() == 2 );
7485 assert( ! reloc->r_pcrel() );
7486
7487 const macho_section<P>* sect = this->machoSection();
7488 Parser<x86>::SourceLocation src;
7489 uint32_t srcAddr = sect->addr() + reloc->r_address();
7490 src.atom = this->findAtomByAddress(srcAddr);
7491 src.offsetInAtom = srcAddr - src.atom->objectAddress();
7492 if ( src.offsetInAtom == 4 ) {
7493 Parser<x86>::TargetDesc stringTarget;
7494 const uint8_t* fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
7495 uint32_t contentValue = LittleEndian::get32(*((uint32_t*)fixUpPtr));
7496 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), stringTarget);
7497
7498 assert(stringTarget.atom != NULL);
7499 assert(stringTarget.atom->contentType() == ld::Atom::typeCString);
7500 const char* superClassBaseName = (char*)stringTarget.atom->rawContentPointer();
7501 char* superClassName = new char[strlen(superClassBaseName) + 20];
7502 strcpy(superClassName, ".objc_class_name_");
7503 strcat(superClassName, superClassBaseName);
7504
7505 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindSetTargetAddress, false, superClassName);
7506 }
7507 }
7508 // inherited
7509 return FixedSizeSection<x86>::addRelocFixup(parser, reloc);
7510 }
7511
7512
7513
7514 template <typename A>
7515 bool Objc1ClassReferences<A>::addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>* reloc)
7516 {
7517 // inherited
7518 PointerToCStringSection<A>::addRelocFixup(parser, reloc);
7519
7520 assert(0 && "needs template specialization");
7521 return false;
7522 }
7523
7524
7525
7526 template <>
7527 bool Objc1ClassReferences<x86>::addRelocFixup(class Parser<x86>& parser, const macho_relocation_info<x86::P>* reloc)
7528 {
7529 // add implict class refs, fixups not usable yet, so look at relocations
7530 assert( (reloc->r_address() & R_SCATTERED) == 0 );
7531 assert( reloc->r_type() == GENERIC_RELOC_VANILLA );
7532 assert( reloc->r_length() == 2 );
7533 assert( ! reloc->r_pcrel() );
7534
7535 const macho_section<P>* sect = this->machoSection();
7536 Parser<x86>::SourceLocation src;
7537 uint32_t srcAddr = sect->addr() + reloc->r_address();
7538 src.atom = this->findAtomByAddress(srcAddr);
7539 src.offsetInAtom = srcAddr - src.atom->objectAddress();
7540 Parser<x86>::TargetDesc stringTarget;
7541 const uint8_t* fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
7542 uint32_t contentValue = LittleEndian::get32(*((uint32_t*)fixUpPtr));
7543 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), stringTarget);
7544
7545 assert(stringTarget.atom != NULL);
7546 assert(stringTarget.atom->contentType() == ld::Atom::typeCString);
7547 const char* baseClassName = (char*)stringTarget.atom->rawContentPointer();
7548 char* objcClassName = new char[strlen(baseClassName) + 20];
7549 strcpy(objcClassName, ".objc_class_name_");
7550 strcat(objcClassName, baseClassName);
7551
7552 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindSetTargetAddress, false, objcClassName);
7553
7554 // inherited
7555 return PointerToCStringSection<x86>::addRelocFixup(parser, reloc);
7556 }
7557
7558 #if SUPPORT_ARCH_arm64
7559 template <>
7560 void Section<arm64>::addLOH(class Parser<arm64>& parser, int kind, int count, const uint64_t addrs[]) {
7561 switch (kind) {
7562 case LOH_ARM64_ADRP_ADRP:
7563 case LOH_ARM64_ADRP_LDR:
7564 case LOH_ARM64_ADRP_ADD:
7565 case LOH_ARM64_ADRP_LDR_GOT:
7566 if ( count != 2 )
7567 warning("arm64 Linker Optimiztion Hint %d has wrong number of arguments", kind);
7568 break;
7569 case LOH_ARM64_ADRP_ADD_LDR:
7570 case LOH_ARM64_ADRP_LDR_GOT_LDR:
7571 case LOH_ARM64_ADRP_ADD_STR:
7572 case LOH_ARM64_ADRP_LDR_GOT_STR:
7573 if ( count != 3 )
7574 warning("arm64 Linker Optimiztion Hint %d has wrong number of arguments", kind);
7575 }
7576
7577 // pick lowest address in tuple for use as offsetInAtom
7578 uint64_t lowestAddress = addrs[0];
7579 for(int i=1; i < count; ++i) {
7580 if ( addrs[i] < lowestAddress )
7581 lowestAddress = addrs[i];
7582 }
7583 // verify all other address are in same atom
7584 Atom<arm64>* inAtom = parser.findAtomByAddress(lowestAddress);
7585 const uint64_t atomStartAddr = inAtom->objectAddress();
7586 const uint64_t atomEndAddr = atomStartAddr + inAtom->size();
7587 for(int i=0; i < count; ++i) {
7588 if ( (addrs[i] < atomStartAddr) || (addrs[i] >= atomEndAddr) ) {
7589 warning("arm64 Linker Optimiztion Hint addresses are not in same atom: 0x%08llX and 0x%08llX",
7590 lowestAddress, addrs[i]);
7591 return; // skip this LOH
7592 }
7593 if ( (addrs[i] & 0x3) != 0 ) {
7594 warning("arm64 Linker Optimiztion Hint address is not 4-byte aligned: 0x%08llX", addrs[i]);
7595 return; // skip this LOH
7596 }
7597 if ( (addrs[i] - lowestAddress) > 0xFFFF ) {
7598 if ( parser.verboseOptimizationHints() ) {
7599 warning("arm64 Linker Optimiztion Hint addresses are too far apart: 0x%08llX and 0x%08llX",
7600 lowestAddress, addrs[i]);
7601 }
7602 return; // skip this LOH
7603 }
7604 }
7605
7606 // encoded kind, count, and address deltas in 64-bit addend
7607 ld::Fixup::LOH_arm64 extra;
7608 extra.addend = 0;
7609 extra.info.kind = kind;
7610 extra.info.count = count-1;
7611 extra.info.delta1 = (addrs[0] - lowestAddress) >> 2;
7612 extra.info.delta2 = (count > 1) ? ((addrs[1] - lowestAddress) >> 2) : 0;
7613 extra.info.delta3 = (count > 2) ? ((addrs[2] - lowestAddress) >> 2) : 0;
7614 extra.info.delta4 = (count > 3) ? ((addrs[3] - lowestAddress) >> 2) : 0;
7615 typename Parser<arm64>::SourceLocation src(inAtom, lowestAddress- inAtom->objectAddress());
7616 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindLinkerOptimizationHint, extra.addend);
7617 }
7618 #endif
7619
7620
7621 template <typename A>
7622 void Section<A>::addLOH(class Parser<A>& parser, int kind, int count, const uint64_t addrs[]) {
7623
7624 }
7625
7626 template <typename A>
7627 void Section<A>::makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&)
7628 {
7629 const macho_section<P>* sect = this->machoSection();
7630 if ( sect->reloff() + (sect->nreloc() * sizeof(macho_relocation_info<P>)) > parser.fileLength() )
7631 throwf("relocations for section %s/%s extends beyond end of file,", sect->segname(), Section<A>::makeSectionName(sect) );
7632 const macho_relocation_info<P>* relocs = (macho_relocation_info<P>*)(file().fileContent() + sect->reloff());
7633 const uint32_t relocCount = sect->nreloc();
7634 for (uint32_t r = 0; r < relocCount; ++r) {
7635 try {
7636 if ( this->addRelocFixup(parser, &relocs[r]) )
7637 ++r; // skip next
7638 }
7639 catch (const char* msg) {
7640 throwf("in section %s,%s reloc %u: %s", sect->segname(), Section<A>::makeSectionName(sect), r, msg);
7641 }
7642 }
7643
7644 // add follow-on fixups if .o file is missing .subsections_via_symbols
7645 if ( this->addFollowOnFixups() ) {
7646 Atom<A>* end = &_endAtoms[-1];
7647 for(Atom<A>* p = _beginAtoms; p < end; ++p) {
7648 typename Parser<A>::SourceLocation src(p, 0);
7649 Atom<A>* nextAtom = &p[1];
7650 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, nextAtom);
7651 }
7652 }
7653 else if ( this->type() == ld::Section::typeCode ) {
7654 // if FDE broke text not at a symbol, use followOn to keep code together
7655 Atom<A>* end = &_endAtoms[-1];
7656 for(Atom<A>* p = _beginAtoms; p < end; ++p) {
7657 typename Parser<A>::SourceLocation src(p, 0);
7658 Atom<A>* nextAtom = &p[1];
7659 if ( (p->symbolTableInclusion() == ld::Atom::symbolTableIn) && (nextAtom->symbolTableInclusion() == ld::Atom::symbolTableNotIn) ) {
7660 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, nextAtom);
7661 }
7662 }
7663 }
7664 if ( !this->_altEntries.empty() && !this->addFollowOnFixups() ) {
7665 if ( _altEntries.count(_beginAtoms) != 0 )
7666 warning("N_ALT_ENTRY bit set on first atom in section %s/%s", sect->segname(), Section<A>::makeSectionName(sect));
7667
7668 Atom<A>* end = &_endAtoms[-1];
7669 for(Atom<A>* p = _beginAtoms; p < end; ++p) {
7670 Atom<A>* nextAtom = &p[1];
7671 // <rdar://problem/22960070> support alt_entry aliases (alias process already added followOn, don't repeat)
7672 if ( (_altEntries.count(nextAtom) != 0) && (p->_objAddress != nextAtom->_objAddress) ) {
7673 typename Parser<A>::SourceLocation src(p, 0);
7674 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, nextAtom);
7675 typename Parser<A>::SourceLocation src2(nextAtom, 0);
7676 parser.addFixup(src2, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinate, p);
7677 }
7678 }
7679 }
7680
7681 // <rdar://problem/9218847> track data-in-code
7682 if ( parser.hasDataInCodeLabels() && (this->type() == ld::Section::typeCode) ) {
7683 for (uint32_t i=0; i < parser.symbolCount(); ++i) {
7684 const macho_nlist<P>& sym = parser.symbolFromIndex(i);
7685 // ignore stabs
7686 if ( (sym.n_type() & N_STAB) != 0 )
7687 continue;
7688 // ignore non-definitions
7689 if ( (sym.n_type() & N_TYPE) != N_SECT )
7690 continue;
7691
7692 // 'L' labels do not denote atom breaks
7693 const char* symbolName = parser.nameFromSymbol(sym);
7694 if ( symbolName[0] == 'L' ) {
7695 if ( strncmp(symbolName, "L$start$", 8) == 0 ) {
7696 ld::Fixup::Kind kind = ld::Fixup::kindNone;
7697 if ( strncmp(&symbolName[8], "data$", 5) == 0 )
7698 kind = ld::Fixup::kindDataInCodeStartData;
7699 else if ( strncmp(&symbolName[8], "code$", 5) == 0 )
7700 kind = ld::Fixup::kindDataInCodeEnd;
7701 else if ( strncmp(&symbolName[8], "jt8$", 4) == 0 )
7702 kind = ld::Fixup::kindDataInCodeStartJT8;
7703 else if ( strncmp(&symbolName[8], "jt16$", 4) == 0 )
7704 kind = ld::Fixup::kindDataInCodeStartJT16;
7705 else if ( strncmp(&symbolName[8], "jt32$", 4) == 0 )
7706 kind = ld::Fixup::kindDataInCodeStartJT32;
7707 else if ( strncmp(&symbolName[8], "jta32$", 4) == 0 )
7708 kind = ld::Fixup::kindDataInCodeStartJTA32;
7709 else
7710 warning("unknown L$start$ label %s in file %s", symbolName, this->file().path());
7711 if ( kind != ld::Fixup::kindNone ) {
7712 Atom<A>* inAtom = parser.findAtomByAddress(sym.n_value());
7713 typename Parser<A>::SourceLocation src(inAtom, sym.n_value() - inAtom->objectAddress());
7714 parser.addFixup(src, ld::Fixup::k1of1, kind);
7715 }
7716 }
7717 }
7718 }
7719 }
7720
7721 // <rdar://problem/11150575> Handle LC_DATA_IN_CODE in object files
7722 if ( this->type() == ld::Section::typeCode ) {
7723 const pint_t startAddr = this->_machOSection->addr();
7724 const pint_t endAddr = startAddr + this->_machOSection->size();
7725 for ( const macho_data_in_code_entry<P>* p = parser.dataInCodeStart(); p != parser.dataInCodeEnd(); ++p ) {
7726 if ( (p->offset() >= startAddr) && (p->offset() < endAddr) ) {
7727 ld::Fixup::Kind kind = ld::Fixup::kindNone;
7728 switch ( p->kind() ) {
7729 case DICE_KIND_DATA:
7730 kind = ld::Fixup::kindDataInCodeStartData;
7731 break;
7732 case DICE_KIND_JUMP_TABLE8:
7733 kind = ld::Fixup::kindDataInCodeStartJT8;
7734 break;
7735 case DICE_KIND_JUMP_TABLE16:
7736 kind = ld::Fixup::kindDataInCodeStartJT16;
7737 break;
7738 case DICE_KIND_JUMP_TABLE32:
7739 kind = ld::Fixup::kindDataInCodeStartJT32;
7740 break;
7741 case DICE_KIND_ABS_JUMP_TABLE32:
7742 kind = ld::Fixup::kindDataInCodeStartJTA32;
7743 break;
7744 default:
7745 kind = ld::Fixup::kindDataInCodeStartData;
7746 warning("uknown LC_DATA_IN_CODE kind (%d) at offset 0x%08X", p->kind(), p->offset());
7747 break;
7748 }
7749 Atom<A>* inAtom = parser.findAtomByAddress(p->offset());
7750 typename Parser<A>::SourceLocation srcStart(inAtom, p->offset() - inAtom->objectAddress());
7751 parser.addFixup(srcStart, ld::Fixup::k1of1, kind);
7752 typename Parser<A>::SourceLocation srcEnd(inAtom, p->offset() + p->length() - inAtom->objectAddress());
7753 parser.addFixup(srcEnd, ld::Fixup::k1of1, ld::Fixup::kindDataInCodeEnd);
7754 }
7755 }
7756 }
7757
7758 // <rdar://problem/11945700> convert linker optimization hints into internal format
7759 if ( this->type() == ld::Section::typeCode && parser.hasOptimizationHints() ) {
7760 const pint_t startAddr = this->_machOSection->addr();
7761 const pint_t endAddr = startAddr + this->_machOSection->size();
7762 for (const uint8_t* p = parser.optimizationHintsStart(); p < parser.optimizationHintsEnd(); ) {
7763 uint64_t addrs[4];
7764 int32_t kind = read_uleb128(&p, parser.optimizationHintsEnd());
7765 if ( kind == 0 ) // padding at end of loh buffer
7766 break;
7767 if ( kind == -1 ) {
7768 warning("malformed uleb128 kind in LC_LINKER_OPTIMIZATION_HINTS");
7769 break;
7770 }
7771 int32_t count = read_uleb128(&p, parser.optimizationHintsEnd());
7772 if ( count == -1 ) {
7773 warning("malformed uleb128 count in LC_LINKER_OPTIMIZATION_HINTS");
7774 break;
7775 }
7776 if ( count > 3 ) {
7777 warning("address count > 3 in LC_LINKER_OPTIMIZATION_HINTS");
7778 break;
7779 }
7780 for (int32_t i=0; i < count; ++i) {
7781 addrs[i] = read_uleb128(&p, parser.optimizationHintsEnd());
7782 }
7783 if ( (startAddr <= addrs[0]) && (addrs[0] < endAddr) ) {
7784 this->addLOH(parser, kind, count, addrs);
7785 //fprintf(stderr, "kind=%d", kind);
7786 //for (int32_t i=0; i < count; ++i) {
7787 // fprintf(stderr, ", addr=0x%08llX", addrs[i]);
7788 //}
7789 //fprintf(stderr, "\n");
7790 }
7791 }
7792 }
7793
7794
7795 // add follow-on fixups for aliases
7796 if ( _hasAliases ) {
7797 for(Atom<A>* p = _beginAtoms; p < _endAtoms; ++p) {
7798 if ( p->isAlias() && ! this->addFollowOnFixups() ) {
7799 Atom<A>* targetOfAlias = &p[1];
7800 assert(p < &_endAtoms[-1]);
7801 assert(p->_objAddress == targetOfAlias->_objAddress);
7802 typename Parser<A>::SourceLocation src(p, 0);
7803 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, targetOfAlias);
7804 }
7805 }
7806 }
7807 }
7808
7809
7810
7811 //
7812 // main function used by linker to instantiate ld::Files
7813 //
7814 ld::relocatable::File* parse(const uint8_t* fileContent, uint64_t fileLength,
7815 const char* path, time_t modTime, ld::File::Ordinal ordinal, const ParserOptions& opts)
7816 {
7817 switch ( opts.architecture ) {
7818 #if SUPPORT_ARCH_x86_64
7819 case CPU_TYPE_X86_64:
7820 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) )
7821 return mach_o::relocatable::Parser<x86_64>::parse(fileContent, fileLength, path, modTime, ordinal, opts);
7822 break;
7823 #endif
7824 #if SUPPORT_ARCH_i386
7825 case CPU_TYPE_I386:
7826 if ( mach_o::relocatable::Parser<x86>::validFile(fileContent) )
7827 return mach_o::relocatable::Parser<x86>::parse(fileContent, fileLength, path, modTime, ordinal, opts);
7828 break;
7829 #endif
7830 #if SUPPORT_ARCH_arm_any
7831 case CPU_TYPE_ARM:
7832 if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, opts.objSubtypeMustMatch, opts.subType) )
7833 return mach_o::relocatable::Parser<arm>::parse(fileContent, fileLength, path, modTime, ordinal, opts);
7834 break;
7835 #endif
7836 #if SUPPORT_ARCH_arm64
7837 case CPU_TYPE_ARM64:
7838 if ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, opts.objSubtypeMustMatch, opts.subType) )
7839 return mach_o::relocatable::Parser<arm64>::parse(fileContent, fileLength, path, modTime, ordinal, opts);
7840 break;
7841 #endif
7842 }
7843 return NULL;
7844 }
7845
7846 //
7847 // used by archive reader to validate member object file
7848 //
7849 bool isObjectFile(const uint8_t* fileContent, uint64_t fileLength, const ParserOptions& opts)
7850 {
7851 switch ( opts.architecture ) {
7852 case CPU_TYPE_X86_64:
7853 return ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) );
7854 case CPU_TYPE_I386:
7855 return ( mach_o::relocatable::Parser<x86>::validFile(fileContent) );
7856 case CPU_TYPE_ARM:
7857 return ( mach_o::relocatable::Parser<arm>::validFile(fileContent, opts.objSubtypeMustMatch, opts.subType) );
7858 case CPU_TYPE_ARM64:
7859 return ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, opts.objSubtypeMustMatch, opts.subType) );
7860 }
7861 return false;
7862 }
7863
7864 //
7865 // used by linker to infer architecture when no -arch is on command line
7866 //
7867 bool isObjectFile(const uint8_t* fileContent, cpu_type_t* result, cpu_subtype_t* subResult, Options::Platform* platform)
7868 {
7869 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) ) {
7870 *result = CPU_TYPE_X86_64;
7871 const macho_header<Pointer64<LittleEndian> >* header = (const macho_header<Pointer64<LittleEndian> >*)fileContent;
7872 *subResult = header->cpusubtype();
7873 *platform = Parser<x86_64>::findPlatform(header);
7874 return true;
7875 }
7876 if ( mach_o::relocatable::Parser<x86>::validFile(fileContent) ) {
7877 const macho_header<Pointer32<LittleEndian> >* header = (const macho_header<Pointer32<LittleEndian> >*)fileContent;
7878 *result = CPU_TYPE_I386;
7879 *subResult = CPU_SUBTYPE_X86_ALL;
7880 *platform = Parser<x86>::findPlatform(header);
7881 return true;
7882 }
7883 if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, false, 0) ) {
7884 const macho_header<Pointer32<LittleEndian> >* header = (const macho_header<Pointer32<LittleEndian> >*)fileContent;
7885 *result = CPU_TYPE_ARM;
7886 *subResult = header->cpusubtype();
7887 *platform = Parser<arm>::findPlatform(header);
7888 return true;
7889 }
7890 if ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, false, 0) ) {
7891 const macho_header<Pointer64<LittleEndian> >* header = (const macho_header<Pointer64<LittleEndian> >*)fileContent;
7892 *result = CPU_TYPE_ARM64;
7893 *subResult = CPU_SUBTYPE_ARM64_ALL;
7894 *platform = Parser<arm64>::findPlatform(header);
7895 return true;
7896 }
7897 return false;
7898 }
7899
7900 //
7901 // used by linker is error messages to describe bad .o file
7902 //
7903 const char* archName(const uint8_t* fileContent)
7904 {
7905 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) ) {
7906 return mach_o::relocatable::Parser<x86_64>::fileKind(fileContent);
7907 }
7908 if ( mach_o::relocatable::Parser<x86>::validFile(fileContent) ) {
7909 return mach_o::relocatable::Parser<x86>::fileKind(fileContent);
7910 }
7911 if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, false, 0) ) {
7912 return mach_o::relocatable::Parser<arm>::fileKind(fileContent);
7913 }
7914 return NULL;
7915 }
7916
7917 //
7918 // Used by archive reader when -ObjC option is specified
7919 //
7920 bool hasObjC2Categories(const uint8_t* fileContent)
7921 {
7922 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) ) {
7923 return mach_o::relocatable::Parser<x86_64>::hasObjC2Categories(fileContent);
7924 }
7925 else if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, false, 0) ) {
7926 return mach_o::relocatable::Parser<arm>::hasObjC2Categories(fileContent);
7927 }
7928 else if ( mach_o::relocatable::Parser<x86>::validFile(fileContent, false, 0) ) {
7929 return mach_o::relocatable::Parser<x86>::hasObjC2Categories(fileContent);
7930 }
7931 #if SUPPORT_ARCH_arm64
7932 else if ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, false, 0) ) {
7933 return mach_o::relocatable::Parser<arm64>::hasObjC2Categories(fileContent);
7934 }
7935 #endif
7936 return false;
7937 }
7938
7939 //
7940 // Used by archive reader when -ObjC option is specified
7941 //
7942 bool hasObjC1Categories(const uint8_t* fileContent)
7943 {
7944 if ( mach_o::relocatable::Parser<x86>::validFile(fileContent, false, 0) ) {
7945 return mach_o::relocatable::Parser<x86>::hasObjC1Categories(fileContent);
7946 }
7947 return false;
7948 }
7949
7950 //
7951 // Used by bitcode obfuscator to get a list of non local symbols from object file
7952 //
7953 bool getNonLocalSymbols(const uint8_t* fileContent, std::vector<const char*> &syms)
7954 {
7955 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) ) {
7956 return mach_o::relocatable::Parser<x86_64>::getNonLocalSymbols(fileContent, syms);
7957 }
7958 else if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, false, 0) ) {
7959 return mach_o::relocatable::Parser<arm>::getNonLocalSymbols(fileContent, syms);
7960 }
7961 else if ( mach_o::relocatable::Parser<x86>::validFile(fileContent, false, 0) ) {
7962 return mach_o::relocatable::Parser<x86>::getNonLocalSymbols(fileContent, syms);
7963 }
7964 else if ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, false, 0) ) {
7965 return mach_o::relocatable::Parser<arm64>::getNonLocalSymbols(fileContent, syms);
7966 }
7967 return false;
7968 }
7969
7970
7971
7972 } // namespace relocatable
7973 } // namespace mach_o
7974
7975
mirror of ld64